Bulletin of Computer Science Research

Innovation in Digital Thesis Supervision Systems Using the Design Thingking Method

Anggreyni Cristine Putricia Br. Surbakti — 2026-06-07

The thesis supervision process represents a critical academic milestone for undergraduate students; however, in many Indonesian higher education institutions, it continues to operate conventionally and inefficiently due to fragmented communication, unstructured revision management, and a lack of organized inter-session support. This study designs and evaluates the BimbOl mobile application prototype as a digital thesis supervision support system using the Design Thinking methodology through five sequential stages: empathize, define, ideate, prototype, and test. A preliminary survey of 57 respondents (45 students and 12 lecturers) identified four dominant barriers: difficulty in scheduling coordination (73.1% of students), unstructured revision management (57.8% of students), absence of proactive notifications (57.9% of students), and loss of inter-session discussion context (53.3% of students). The BimbOl prototype integrates six core features across 30 interactive screens, encompassing collaborative scheduling, digital revision markup, automated notifications, a progress dashboard, structured revision history, and a four-mode AI Consultation feature as an independent inter-session support tool not found in comparable systems. Usability evaluation using the System Usability Scale (SUS) involved 15 respondents (12 students and 3 lecturers) and yielded a combined score of 80.8, surpassing the Grade A threshold (80.3), confirming Excellent/Acceptable usability. The mean student score of 77.9 (Good/Acceptable) and the mean lecturer score of 90.0 (Excellent/Best Imaginable) confirm that the design is effective across both user groups. Unlike prior studies that addressed thesis supervision digitalization through isolated features or interface-level redesigns, this study contributes a unified, human-centered mobile supervision ecosystem that simultaneously resolves scheduling, revision tracking, notification, and AI-assisted inter-session consultation, a combination not previously documented in the Indonesian higher education context, thereby providing an empirical and design foundation for next-generation thesis supervision systems.

Klasifikasi Tingkat Pemahaman Siswa Kelas VI Sekolah Dasar terhadap Perangkat Keras Komputer Menggunakan Metode Decision Tree

Suchi Azzahro Syarif — 2026-06-07

The development of information technology in education requires students to have a basic understanding of computer hardware from an early age. However, the level of students’ understanding of computer hardware still varies, especially at the elementary school level. This condition can affect students’ ability to understand the use of technology more effectively in computer-based learning processes. This study aims to classify the level of understanding of sixth grade elementary school students regarding computer hardware using the Decision Tree method. The research data were obtained through a questionnaire consisting of 25 questions related to computer hardware. Each student’s answer was assigned points based on its correctness level, then the total score was calculated and converted into a 0–100 scale before being categorized into three classes, namely High Understanding, Moderate Understanding, and Low Understanding based on score ranges adjusted to the distribution of the research data. The data show that there are 30 students in the High Understanding category, 18 students in the Moderate Understanding category, and 6 students in the Low Understanding category. The classification process was carried out using the Decision Tree method with 80% training data and 20% testing data. The model achieved an accuracy of 45% on the test data. The result indicates that the model is not yet optimal in performing balanced classification across all categories of student understanding. The findings of this study contribute to the application of the Decision Tree classification method in elementary education, particularly in identifying students’ understanding of computer hardware based on questionnaire data.

Implementasi Algoritma C4.5 Untuk Klasifikasi Pengenalan Warna Dasar di Taman Kanak-Kanak

Anandaya Difi Dzulardi Kalimanti — 2026-06-07

Differences in early childhood ability to recognize basic colors at TK Negeri 01 Barong Tongkok indicate the need for a structured evaluation system to ensure objective assessment. The classification of these abilities is carried out by applying the C4.5 algorithm within a quantitative experimental framework. Data are collected through observations and color recognition tests involving 35 children as respondents, then processed using predefined attributes to construct a classification model. The analysis results group children’s abilities into three categories: Sangat Mengenal (High), Mengenal (Moderate), and Cukup Mengenal (Low). The experimental results indicate that the C4.5 algorithm is highly effective and stable, achieving an average classification accuracy of 85.71% through 5-Fold Cross-Validation. Furthermore, the resulting decision tree provides an intuitive and transparent structure that assists educators in interpreting evaluation outcomes and understanding the dominant variables that determine student learning success more clearly than black-box models. The primary contribution of this study lies in the provision of a data-driven evaluation model that generates empirically measurable decision rules (if-then rules), while simultaneously serving as a methodological bridge to create differentiated learning strategies at the early childhood education (PAUD) level. Consequently, the implementation of the C4.5 algorithm represents a strategic, efficient, and scientifically accountable alternative for enhancing pedagogical effectiveness and cognitive monitoring in early childhood education.

Perancangan Aplikasi FinanSmart dengan Financial Health Score dan Family Link untuk Manajemen Keuangan Mahasiswa

Natasha Patricia Nainggolan — 2026-06-07

Students often face difficulties in systematically managing their personal finances due to limited financial literacy and the absence of applications specifically designed for their needs. This condition can lead to consumptive behavior, inability to save, and prolonged financial dependency. This study aims to design and develop a web-based student financial management application named FinanSmart using the Research and Development (R&D) method with the Waterfall development model. The main innovations of this system include a Financial Health Score (FHS) mechanism calculated automatically each month as a quantitative indicator of students' financial health, and a Family Link feature that allows parents to monitor their child's financial condition in real-time while full privacy control remains with the student. Additionally, the system is equipped with spending pattern analysis, two-level budget notifications, daily reminders, and gamification elements to encourage consistent financial management. Functional testing using Black Box Testing on 18 test scenarios demonstrated a 100% success rate, proving that all features operate according to functional specifications. This system is expected to measurably improve students' financial awareness and behavior through personalized financial health score feedback and family-based monitoring features.

Optimasi Random Forest Menggunakan Domestic Cattle Optimization Algorithm (DCOA) Untuk Diagnosa Somnipati

Muhamad Toriq Aziz Firdaus — 2026-06-07

Sleep disorders such as insomnia and sleep apnea, collectively referred to as somnipathy, are health conditions that can significantly reduce quality of life and are associated with various chronic diseases. However, conventional diagnostic methods such as polysomnography (PSG) have limitations in terms of cost, time, and accessibility. Therefore, this study proposes a machine learning–based approach to classify sleep disorders using a combination of Random Forest integrated with the SMOTE and the DCOA. The dataset used in this study is the Sleep Health and Lifestyle Dataset, which consists of 374 records with 13 features representing individuals’ physiological conditions and lifestyle factors. The issue of data imbalance is addressed using SMOTE, while hyperparameter optimization is performed using DCOA to enhance model performance. The findings indicate that the proposed model achieves an accuracy of 97.33%, precision of 97.63%, recall of 97.33%, and an F1-score of 97.32%. These results demonstrate a significant improvement compared to previous studies using the same dataset. Therefore, the proposed approach proves to be effective in improving sleep disorder classification performance and has strong potential to be implemented as an optimal and accurate data-driven decision support system for diagnosis. However, considering the relatively small dataset size, there is a potential risk of overfitting, which necessitates careful evaluation to ensure model generalization.

Simulasi dan Uji Fungsional Dashboard Monitoring Kualitas Air Kolam Ikan Berbasis MQTT

Anjelita Trully Carolina Losung — 2026-06-07

Water quality plays a crucial role in fish farming because changes in pH, temperature, water content, turbidity, and total dissolved solids (TDS) can affect the condition and health of fish. Manual monitoring still has limitations because data is not available continuously, and early warnings are difficult to provide when water conditions begin to exceed safe limits. This study aims to develop and functionally test an Internet of Things (IoT)-based fish pond water quality monitoring dashboard using the MQTT protocol. Testing was conducted in a simulated environment to evaluate data transmission, data storage, dashboard visualization, status classification, and alert logging. Test results show that the system is capable of receiving and storing 204 sensor data records and generating 248 alert data entries in the `alert_log` table. The dashboard successfully displays the statuses NORMAL, WARNING, DANGER, and CRITICAL in accordance with the defined threshold rules. This research contributes to the development of an initial prototype of a simulation-based water quality monitoring system that can be used to verify data flow integration and dashboard responses before implementation on real devices. This research is not intended as a validation of physical sensors or an evaluation of field network performance, but rather as an initial test of the simulation system’s functionality.

Clustering State Electricity Company (PLN) Customer Electricity Consumption Patterns Using K-Means for Operational Efficiency

Anang Putranto — 2026-06-07

Electricity consumption patterns among PLN customers show diverse characteristics that require data-driven analysis to support accurate operational planning and service strategies. Differences in installed power, operating hours, and electricity usage intensity can indicate variations in customer behavior, load demand, and service needs. This study aims to analyze electricity consumption patterns of PLN customers in the South Semarang area using the K-Means Clustering method. The dataset consists of 6,622 active customers, with variables including installed power, operating hours, and average electricity consumption over the last eight months. The research stages include identifying the need for consumption pattern analysis, preparing data, cleaning incomplete or inconsistent values, normalizing variables, modeling with K-Means, and interpreting cluster results. The optimal number of clusters was determined using the Elbow Method and Silhouette Score to ensure meaningful grouping and good separation quality. The results show that the optimal number of clusters is three. The first cluster represents low consumption patterns with 1,881 customers (28.4%). The second cluster represents medium consumption patterns with 2,927 customers (44.2%). The third cluster represents high consumption patterns with 1,814 customers (27.4%). The WCSS value of 498.67 and Silhouette Score of 0.68 indicate good clustering performance and practical usefulness for decision-making by PLN in service quality improvement.

Perancangan Backend Portal Informasi Komite Olahraga Nasional Indonesia Denpasar Berbasis MVC

Hisyam Akmal Maulana — 2026-06-07

The rapid development of information technology has encouraged many organizations to implement website-based information systems to improve the effectiveness of data management and digital information distribution. KONI Denpasar, as an organization engaged in regional sports development, requires a backend system capable of supporting integrated information management because the previous data management process was still carried out conventionally and was not centralized. This condition caused the dissemination of organizational information to become less optimal and made administrative processes and long-term system development more difficult. Several previous studies have discussed the development of systems using the CodeIgniter framework; however, most of these studies focused on academic systems, administrative services, and sales systems, while research related to backend information portals for sports organizations remains limited. Therefore, this study aims to design and develop a website-based backend information portal for KONI Denpasar using the CodeIgniter 4 framework with the Model View Controller (MVC) architectural approach. The system development method applied in this research is the Systems Development Life Cycle (SDLC) using the Waterfall model, which consists of requirements analysis, system design, implementation, testing, and maintenance stages. The developed backend system includes features for managing news, event agendas, documentation galleries, user data, and administrator access rights to support more structured information management. System testing was conducted using the Black Box Testing method to ensure that all features functioned according to user requirements. The results of this study indicate that the backend information portal for KONI Denpasar was successfully developed and that all system features operated properly based on the testing results. This research contributes to the development of an integrated MVC-based backend information portal for sports organizations that can support information management in a more effective, structured, and easily maintainable manner in the future.

User Experience Evaluation of the SEKAWAN V3 Website Using the User Experience Questionnaire (UEQ)

A'rafi Laksmana Dirgantara — 2026-06-07

The SEKAWAN V3 website is used as a thesis management system in the Informatics Study Program at Universitas Islam Indonesia. However, preliminary observations indicate usability issues affecting user interaction. This study aims to assess the user experience of SEKAWAN V3 and analyze the impact of a UI/UX redesign based on user feedback. The study applies the User Experience Questionnaire (UEQ) supported by open-ended responses. A total of 23 participants were involved, with 17 valid responses in the initial evaluation and 22 valid responses in the re-evaluation after data consistency filtering. The initial results reveal issues related to navigation clarity, feature understanding, and interface structure. Based on these findings, a redesigned prototype was developed and evaluated. The results indicate improvements across all UEQ dimensions, particularly in stimulation, attractiveness, and perspicuity, suggesting better engagement, visual appeal, and ease of use. Qualitative findings support these results, showing that users perceive the system as clearer, more structured, and more intuitive. This study demonstrates that combining quantitative and qualitative approaches can support data-driven improvements in system design.

Implementasi Sistem Informasi Nilai Akademik dan Raport Berbasis Web Menggunakan Metode Object Oriented Analysis and Design

Subarkah Abdullah — 2026-06-07

Managing academic grade data and creating report cards are crucial activities in educational administration, but the current process is still carried out conventionally using spreadsheet applications. This method of working causes various problems such as long processing time, high risk of calculation errors, and difficulties in storing and tracking historical data. This study aims to design, build, and implement a web-based information system that is able to automate grade processing and report card creation to be faster, more accurate, and more integrated. The method applied in developing this system is Object Oriented Analysis and Design (OOAD) with modeling using the Unified Modeling Language (UML), and implementation using the PHP programming language and MySQL database. The main contribution of this research lies in the development of a web-based academic information system specifically designed to meet the characteristics of vocational high school administration, by applying the Object-Oriented Analysis and Design (OOAD) approach and structured UML modeling. The resulting system not only automates grade processing and report card printing, but also provides a modular, integrated, and easily developed software architecture to support the ongoing digitalization of educational administration.

An Agile-Scrum Approach to Enhancing Urban Transit through a Mobile-Based Motorcycle Parking System with OCR Integration

Abadi Suryo Setiyo — 2026-06-07

This study discusses the development of a mobile-based motorcycle storage application using the Agile Scrum method in the 359 Parking service in the Pondok Cina Station area. The application was developed to help manage vehicle storage digitally through storage location search features, vehicle reservations, and vehicle identification using Optical Character Recognition (OCR) technology. The system is built using Flutter and Firebase with the implementation of Google ML Kit OCR to help with the automatic scanning of vehicle license plates. Data collection was carried out through literature studies, interviews, and numerical scale forms on customers and motorcycle storage managers. System testing is carried out using the Black Box Testing method on the main features of the application. The test results show that the login, reservation, location search, and vehicle license plate scanning features can run according to the functional needs of the system. The implementation of OCR also helps reduce the manual recording process of vehicles, especially in adequate lighting conditions. In addition, user evaluation results show that the app helps simplify the process of finding a storage location and managing a vehicle. This study shows that the application of mobile technology, location services, and OCR can support the digitization of motorcycle storage services to be more effective and integrated.

Klasifikasi Penyakit Pneumonia Menggunakan Regresi Logistik, SVM, dan Fitur Deep Learning

Ahmad Bagus Muzakki — 2026-06-20

This study evaluates the integration of deep learning-based feature extraction with conventional machine learning algorithms for pneumonia disease classification from chest X-ray images. Two pre-trained models, Inception V3 and SqueezeNet, are used as feature extractors due to their ability to generate effective feature representations from medical images. Inception V3 was chosen because it is able to capture complex visual patterns, while SqueezeNet offers computational efficiency with a smaller number of parameters. Meanwhile, Logistic Regression and Support Vector Machine (SVM) are used as classification algorithms due to their ability to handle high-dimensional data extracted from deep learning features. The dataset used consists of two categories, namely normal images and pneumonia images, with the entire analysis process carried out using Orange Data Mining. The experimental results show that the combination of Inception V3 and SVM provides the best performance with an Area Under Curve (AUC) of 0.731, Classification Accuracy (CA) of 0.835, F1-score of 0.805, Precision of 0.810, Recall of 0.835, and Matthews Correlation Coefficient (MCC) of 0.321. Meanwhile, the combination of SqueezeNet and Logistic Regression produces a CA of 0.771, an F1-score of 0.751, and an MCC of 0.118, showing quite competitive performance although still below Inception V3 and SVM. The results show that the quality of feature embedding generated by the deep learning model has a significant influence on classification performance. The integration of transfer learning and conventional machine learning has been proven to improve the accuracy of pneumonia detection and has the potential to support the development of an efficient and accurate artificial intelligence-based diagnostic system.

Transformasi Pencatatan Keuangan Melalui Implementasi Odoo POS dan Accounting Berbasis RAD Pada UMKM

Roro Mawar Amalia — 2026-06-20

This study discusses the integrated implementation of Odoo Point of Sale (POS) and Accounting modules at Tahu Bandung Sari Mawar, a small and medium enterprise (SME) that previously managed financial records manually using notebooks. The main issues identified were unstructured transaction recording, difficulties in retrieving past data, poorly documented expense transactions, and the absence of monthly financial reports. This research employed a qualitative case study approach using the Rapid Application Development (RAD) method adapted for ready-made software implementation. In this study, the construction phase focused on system configuration, workflow adjustment, module integration, and testing instead of software development from scratch. Data were collected through observation, interviews, and documentation. System testing used Black Box Testing and User Acceptance Testing (UAT), while evaluation was conducted through qualitative comparative analysis. Black Box Testing results showed that all 7 testing scenarios were successfully executed. UAT results obtained a score of 42 out of 50 (84%), indicating good user acceptance. The comparative evaluation showed that transaction recording time decreased from approximately 5 minutes to 2 minutes after implementation. In addition, Odoo enabled automatic integration between POS and Accounting modules, more detailed transaction recording, and the generation of financial reports from recorded transactions.

Implementasi Frontend Sistem Pelaporan Tugas Harian Berbasis Web dengan Pendekatan Gamifikasi Menggunakan Vue.js

Riza Al Hambra — 2026-06-20

The phenomenon of students' lack of discipline in completing and reporting daily tasks is an academic problem that impacts learning achievement. Monotonous manual reporting systems often reduce intrinsic motivation and student participation rates. This study aims to design and implement a web-based daily task reporting system frontend with a gamification approach using the Vue.js framework, and analyze its impact on improving discipline. The research location is at Universitas Muhammadiyah Karanganyar involving 10 active students as trial respondents. The research method used is a qualitative approach with an implementative case study type. The system interface was developed using Vue.js 3, Pinia as state management, and Vue Router. Data collection techniques included participatory observation, in-depth interviews, and User Acceptance Test (UAT) questionnaires with a 1 to 5 Likert scale. Data testing was analyzed using the Miles and Huberman model consisting of data reduction, data display, and drawing conclusions. Interim research results show the functional success rate of the frontend reached 88 percent in the localhost environment. The UAT questionnaire obtained an overall average score of 4.21 which is classified as very good. Gamification features became the most preferred element by 60 percent of respondents, which effectively encouraged students' disciplined behavior in submitting assignments on time.

Implementation of PROMETHEE Method in Decision Support System for Student Competency Competition Participant Selection

Muhamad Rezka Al Anshori — 2026-06-20

The Student Competency Competition (Lomba Kompetensi Siswa/LKS) is a prestigious national-level event requiring vocational schools to select participants in an objective and measurable manner. However, the selection process at SMKN 1 Tangerang Selatan still relies on subjective teacher judgment and lacks comprehensive evaluation criteria, risking suboptimal decisions that may result in selected students being insufficiently competent for the competition. This research aims to design and implement a web-based Decision Support System (DSS) using the PROMETHEE (Preference Ranking Organization Method for Enrichment Evaluation) method to support objective, structured, and transparent LKS participant selection. Three evaluation criteria were defined based on field interviews: Practicum Subject (K1), Logic Subject (K2), and Attendance (K3), each processed through pairwise preference comparisons using linear and usual preference functions with equal criterion weights. The system was developed using JavaScript (Node.js) and MongoDB, supporting four user roles with distinct access rights. System correctness was verified through Black Box Testing across 16 functional scenarios and White Box Testing of four algorithmic modules using Cyclomatic Complexity analysis, all of which passed successfully. The ranking results identified A3 as the top-recommended candidate with the highest Net Flow value of +0.2083, demonstrating the system's ability to surface competency dimensions overlooked by manual evaluation. Based on a questionnaire involving 20 respondents, the system achieved a user satisfaction score of 87.6%, categorized as very strong. These findings confirm that the PROMETHEE-based DSS is a reliable, practical, and deployment-ready tool for supporting participant selection in vocational school environments. The findings of this study offer a replicable methodological framework for vocational institutions seeking to standardize competition-based participant selection through multi-criteria decision support.

Pengembangan Aplikasi Smart Transportation Berbasis Real-Time Tracking dan Pembayaran Digital untuk Bus Listrik Menggunakan Metode Research and Development (R&D)

Dolly Brams Matondang — 2026-06-20

The development of digital technology has encouraged innovation in public transportation services through the concept of smart transportation. However, electric bus services still face limitations in providing real-time bus location information and have not yet integrated digital payment systems within a single platform. These conditions make it difficult for users to obtain arrival time information and perform payment transactions efficiently. This study aims to develop a mobile-based smart transportation application that integrates real-time tracking, Estimated Time of Arrival (ETA), route and bus stop information, and digital payment services through mobile banking, e-wallets, and QR code. The Research and Development (R&D) method was applied because it supports a systematic system development process starting from needs analysis, system design, prototype development, and usability testing. The contribution of this research lies in the development of an integrated application to facilitate users in accessing electric bus services more effectively and efficiently, considering that there is currently no application that completely integrates these features into a single platform. Usability testing using the System Usability Scale (SUS) method involving 30 respondents produced an average score of 87.3, which falls into the Excellent, Grade A, and Acceptable categories. These results indicate that the application has a very good level of usability and is feasible for further development.

Influence of Imbalanced Data on Text Classification Using Recurrent Neural Network

Rina Septiriana — 2026-06-20

Recurrent Neural Networks (RNNs) such as LSTM and GRU are designed for sequential data. However, their performance in emotion detection is often compromised by class imbalance. This study compares LSTM and GRU architectures for classifying emotional states using a dataset of 4,386 Indonesian tweets. The dataset exhibits a mild imbalance (approximately 1.7:1) across five classes: Anger, Happy, Sadness, Love, and Fear. However, the effectiveness of these models is often hindered by class imbalance in datasets, which biases predictions toward majority classes and compromises the reliability of standard metrics. This study aims to systematically evaluate the comparison of LSTM and GRU architectures in processing imbalanced Indonesian emotional tweet data. The methodology involves evaluating these models across various resampling techniques, including Random Oversampling, SMOTE, and Near-Miss. Key findings reveal that LSTM consistently outperforms GRU in capturing complex emotional patterns. Specifically, the LSTM model combined with Random Oversampling emerged as the most robust configuration, achieving a Macro-F1 score of 71% and an accuracy of 73%. While Random Oversampling effectively enhanced minority class recognition without overfitting, SMOTE and Near-Miss introduced significant performance trade-offs. These results provide actionable insights for selecting optimal architectures and resampling strategies to mitigate imbalance-related biases in sequential classification tasks.

Analisis Penerimaan Sistem E-Commerce Smartschool Menggunakan Model Unified Theory of Acceptance and Use of Technology

Akram Farrasanto — 2026-06-20

The conventional school uniform purchasing process still faces various obstacles, such as limited service time, inefficient transaction recording, and minimal real-time transaction documentation. Therefore, to overcome these problems, SMK Bina Patriot implemented the SmartSchool e-commerce system as a transaction medium for purchasing school uniforms. This study aims to analyze user acceptance of the SmartSchool e-commerce system using the Unified Theory of Acceptance and Use of Technology (UTAUT) model. The UTAUT model is used to test the effect of Performance Expectancy, Effort Expectancy, Social Influence, and Facilitating Conditions on Behavioral Intention in using the system. This study uses a quantitative approach by using data collection techniques through distributing questionnaires to 55 students as respondents in the study. Then the resulting data will be analyzed using multiple linear regression to determine the effect of each variable on the intention to use the system. The results of this study also show that Performance Expectancy has a significant effect on Behavioral Intention with a significance value of 0.037 (p < 0.05), while Social Influence also has a significant effect with a significance value of 0.002 (p < 0.05). Conversely, Effort Expectancy did not significantly influence Behavioral Intention, with a significance value of 0.667 (p > 0.05). Therefore, it can be seen that the variable with the most dominant influence is Social Influence, with a regression coefficient of 0.4882. This study contributes to extending the application of the UTAUT model within the context of school e-commerce systems at the secondary education level, which remains relatively underexplored in existing research. Furthermore, the findings of this study can serve as a reference for schools in designing implementation strategies and developing e-commerce systems that better align with users' needs and expectations

Perancangan Sistem Informasi Pengelolaan Pakan Ternak Berbasis Web Menggunakan Metode Waterfall

Hizki Alawiyah — 2026-06-20

Animal feed management plays a crucial role in supporting the smooth operation of livestock farms, particularly in government agencies involved in livestock development and breeding. The Margawati Sheep and Goat Livestock Breeding Development Center (BPPTDK) still employs manual feed data management processes, from recording feed types, stock management, to report preparation. This situation results in a suboptimal data management process due to the potential for recording errors, delays in information delivery, and difficulties in accurately monitoring feed stocks. This study aims to design a Web-Based Animal Feed Management Information System to support a more effective, integrated, and computerized feed data management process. The system development method used is the Waterfall method, which includes the stages of needs analysis, system design, implementation, testing, and maintenance. The system design process was carried out using the Unified Modeling Language (UML) in the form of use case diagrams, activity diagrams, and class diagrams. The results of the study indicate that the developed system is able to assist in managing feed type data, controlling feed stock, recording incoming and outgoing feed data, and preparing reports more effectively compared to the previous manual system. Test results using the Black Box Testing method indicate that all the system's main functions operate according to user requirements. The designed system also supports real-time feed stock monitoring and facilitates integrated data management according to operational requirements at the Margawati BPPTDK UPTD. Therefore, the developed information system is expected to improve the effectiveness and efficiency of the livestock feed management process in a sustainable manner.

Applying the Waterfall Method and Temporal Overlap Detection Algorithm in a Cross-Platform Badminton Court Booking Application

Danis Wara Wardana — 2026-06-20

This study directly addresses the specific scheduling bottlenecks and operational inefficiencies at Furry Sport Center Seyegan, Sleman. The conventional booking workflow, dependent on manual physical logbooks and fragmented WhatsApp communications, suffers from severe limitations such as high vulnerability to double bookings and delayed staff responses. The digitalization of this management system is highly urgent to prevent ongoing operational disruptions and potential revenue loss caused by these manual errors. To resolve this, a cross-platform mobile booking application was developed utilizing the Waterfall software development model, engineered with a Flutter and Node.js architecture. The system incorporates a Temporal Overlap Detection Algorithm to systematically ensure conflict-free scheduling. To provide objective evidence of efficiency and eliminate subjective claims, system validation was conducted using the System Usability Scale (SUS). The testing yielded an overall average score of 79.25, placing the application in the acceptable usability category with high user satisfaction. These empirical parameters demonstrate that the application measurably resolves conventional scheduling conflicts and provides a proven, transparent solution for local sports venue management.

Analisis Spasial Temporal Pola Gempa Bumi Menggunakan GIS dan Visualisasi Peta Animasi

Mahreza Rakha Indra Putra — 2026-06-20

Central Aceh Regency is one of the regions in Aceh Province influenced by the tectonic activity of the Sumatra Fault, resulting in relatively high seismic activity. This study aims to analyze the spatial and temporal changes in earthquake patterns in Central Aceh Regency during 2004–2015 using Geographic Information Systems (GIS) and animated map visualization. The data used in this study were obtained from the United States Geological Survey (USGS) and the Meteorology, Climatology, and Geophysics Agency (BMKG) in CSV format, including parameters such as event time, magnitude, depth, latitude, and longitude. The research stages consisted of data preprocessing, data de-duplication, Kernel Density Estimation (KDE) analysis, spatial-temporal analysis, and visualization using the QGIS Temporal Controller. The coordinate system applied in this study was WGS 84 (EPSG:4326). The results indicate a shift in earthquake epicenter distribution from the northwest toward the southeast of Central Aceh during the observation period. In 2004, 2,006 earthquake events were recorded with an average magnitude of 4.49 Mw and an average depth of 84.82 km, while in 2015, 116 events were recorded with an average magnitude of 5.1 Mw and an average depth of 30 km. Heatmap analysis revealed increasing seismic concentration near the active Sumatra Fault zone. The animated map visualization successfully illustrated the temporal dynamics of seismic activity and can support spatial-based disaster mitigation planning. The main contribution of this study is the development of a GIS-based spatial-temporal analysis model that not only presents earthquake point distribution but also dynamically shows changes in seismic concentration through heatmaps and animated visualization. The findings can serve as an initial basis for identifying earthquake-prone zones and supporting spatial data-based disaster mitigation planning.

Implementasi Local-First RAG dengan Hybrid Retrieval IndoBERT dan BM25 untuk Pendukung Keputusan Akademik

Romi Wahyudi Hasibuan — 2026-06-20

The need for rapid and accurate access to academic information for educational institution stakeholders, such as campus management and academic advisors, still relies on manual administrative processes, thereby leading to operational inefficiencies. This study develops a locally hosted, open-source Retrieval-Augmented Generation (RAG) system as an automated solution for academic information services while alleviating administrative burdens. The system integrates the Mistral-7B-Instruct LLM with a hybrid search approach within the Elasticsearch ecosystem, combining IndoBERT-based dense retrieval for narrative academic guideline documents and BM25-based sparse retrieval for structured student data. Evaluation was conducted using ROUGE-1, ROUGE-2, and ROUGE-L metrics against 60 test data points generated by Claude Sonnet 4.6. The system successfully answered 58 out of 60 queries, achieving a ROUGE-L f1-score of 0.29. An asymmetrical pattern was observed, where recall values were consistently higher than precision across all metrics, which indicates the impact of the language generation capacity gap between Mistral 7B and the reference model. An average input prompt length ranging from 1,270 to 1,380 tokens contributed to an average latency of 30 seconds per query, representing the primary contemporary challenge of the system. This research is expected to serve as a baseline for developing open-source RAG systems within Indonesian language domains, specifically in the context of higher education academic administration.

Implementasi Regresi Linier Berganda Untuk Prediksi Harga Mobil Bekas Di Indonesia Berbasis Gradio

M Ridho Alfani — 2026-06-20

The price of a used vehicle depends on various aspects that cause changes in the selling value in the market, such as model, year, transmission, mileage, fuel, tax, mpg, and cc. A common problem in used car transactions is determining prices that are still not fully based on measurable data analysis. The purpose of this study is to design a model to estimate the price of a used car through the multiple linear regression method and implement it in the User Interface. The data used in this study is secondary data obtained from the Kaggle public repository, and collected from several used car buying and selling forums in Pekanbaru and social media platforms such as Facebook that contain vehicle price information. The dataset contains 400 rows of data with a range of car years from 2005 to 2025. The research stages include data preprocessing in the form of categorical variable encoding and data normalization. Data is divided into training data and testing data, followed by the process of model building and model performance assessment. Evaluation is carried out using the Mean Squared Error (MSE), Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), and coefficient of determination (R²) metrics. The model was built using several independent variables, namely model, year, transmission, kilometer, fuel, tax, mpg, and cc, with vehicle price as the dependent variable. Based on the test results, the multiple linear regression method shows the ability to produce used car price estimates and has potential for application in decision support systems. The test results show that the MSE value on the training data is 0.004 and the testing data is 0.010, MAE on the training data is 0.046 and the testing data is 0.071, and RMSE on the training data is 0.062 and 0.100 on the testing data, the coefficient of determination (R²) on the training data is 0.985 and on the testing data is 0.955. The next model is implemented using the Python Gradio library so that users can predict vehicle prices through the User Interface.

Analisis Pengembangan dan Evaluasi Virtual World Bertema Lingkungan Pegunungan Menggunakan Metode Research and Development (R&D)

Robby Firmansyah — 2026-06-20

The development of metaverse technology and virtual worlds has created opportunities for utilizing virtual environments as media for digital exploration, simulation, and interactive learning. However, most previous studies have primarily focused on pedagogical aspects and learning simulations, while research discussing the technical development of terrain modeling, lighting, and mountainous environmental atmospheres on the Roblox Studio platform remains limited. This study aims to develop and evaluate a mountainous environment-themed virtual world using Roblox Studio and to analyze user acceptance of the resulting virtual environment. The research employed the Research and Development (R&D) method, consisting of needs analysis, design, implementation, testing, and evaluation stages. The development process utilized Terrain Editor, Future Lighting, skybox, fog, and Lua scripting features to create an interactive virtual environment. Evaluation was conducted using a Likert-scale questionnaire distributed to 67 respondents. The results showed a feasibility score of 79.88%, which falls into the good category. The highest-rated indicator was the suitability of the virtual world to the mountainous environment concept (4.15), while the lowest-rated indicator was the structured layout of the environment (3.83). These findings indicate that the developed virtual world was able to provide a positive exploration experience for users. The contribution of this research lies in establishing a systematic development framework for a mountainous environment-themed virtual world through the implementation of terrain modeling, lighting and atmospheric configuration, and user experience evaluation using the Research and Development (R&D) approach. In addition, this study provides a practical reference for the development of metaverse-based virtual environments that can be utilized for simulation, digital exploration, and interactive learning applications.

Analisis Kinerja Recursive Feature Elimination pada Support Vector Machine untuk Klasifikasi Penyakit Stroke pada Data Tidak Seimbang

Faridatul Jannah — 2026-06-20

Stroke is a non-communicable disease with high mortality and disability rates, necessitating a classification approach that can facilitate more effective detection. Class imbalance in stroke datasets causes classification models to be biased toward the majority class, resulting in suboptimal classification performance. This study aims to analyze the performance of Recursive Feature Elimination (RFE) in a Support Vector Machine (SVM) model with data imbalance handling using Adaptive Synthetic Sampling (ADASYN) in stroke classification. The dataset used is a secondary dataset from Kaggle consisting of 5109 data points after the preprocessing stage. The modeling process was conducted by testing various data split ratios as well as combinations of kernels and SVM parameters using a 5-fold cross-validation approach. The results show that the best model was obtained with an 80:20 split ratio, a polynomial kernel, and a C parameter of 0.1, yielding an accuracy of 0.75, precision of 0.14, recall of 0.82, an F1-score of 0.24, and an AUC of 0.8245. The application of RFE resulted in improved model performance compared to without RFE, although the magnitude of the improvement was relatively small. The still low precision value indicates that the model still produces many false positives, so the classification challenge on the stroke dataset has not been fully resolved. On the other hand, an AUC value of 0.8245 indicates that the model performs reasonably well in distinguishing between the two classes overall, although its application in a clinical context still requires further refinement.

Evaluasi Efektifitas Optimizer Adam dan SGD pada Klasifikasi Citra Dermoskopi dengan MobileNetV4

Ahmad Naufal — 2026-06-20

Skin disease is one of the most common health problems and requires fast and accurate diagnosis. The limited availability of dermatology specialists and the high subjectivity of conventional diagnosis have driven the development of artificial intelligence-based automatic classification systems. This study aims to compare the performance of the Adam and Stochastic Gradient Descent (SGD) optimizers on the MobileNetV4 architecture for classifying eight classes of skin diseases using the ISIC 2019 dataset. The dataset consists of 23,257 valid dermoscopic images after preprocessing, which includes duplicate image removal, hair artifact elimination using the blackhat morphology method, and an asymmetric sampling strategy in which majority classes were capped at a maximum of 2,000 images while minority classes were augmented to reach the target count, in order to address extreme class imbalance with a ratio of up to 53:1. The model was trained using a three-phase training strategy with gradual unfreezing of the MobileNetV4 backbone initialized with pretrained ImageNet weights. All training configurations were made identical for both optimizers except for the optimization algorithm and learning rate, ensuring a fair comparison. Evaluation results on the test set show that the Adam optimizer achieved an accuracy of 71.07% with a macro F1-score of 0.72, while SGD achieved an accuracy of 58.06% with a macro F1-score of 0.57. Adam outperformed SGD across all eight skin disease classes. The performance difference of 13.01% indicates that Adam's adaptive learning rate mechanism is more effective for dermoscopic datasets with imbalanced class distributions compared to SGD. Nevertheless, it should be noted that Adam requires greater computational memory than SGD due to the storage of first and second moment estimates per parameter, and therefore the computational efficiency trade-off should be considered when deploying the model on resource-constrained devices. This study provides empirical contribution in selecting the optimal optimizer for skin lesion classification based on lightweight architectures.

Analisis Pengelompokan Jenis Anomali Aktivitas Pengguna Pada Log Sistem Informasi Klinik Menggunakan Lof Dan K-Means

Puja M Alca — 2026-06-20

Digital transformation in the healthcare sector has driven the adoption of clinic information systems for computerized management of patient medical records. Sensitive data security is threatened by user behavior deviations, requiring immediate detection mechanisms. This study aims to identify anomalous activity patterns and indicators from user log records, including unusual database operation frequencies, abnormal access times, and suspicious data manipulation patterns.The Local Outlier Factor algorithm functions to systematically calculate the local density score of each data point relative to its nearest neighbors. This method detects user activities that deviate significantly from normal patterns in daily clinic operational systems. The K-Means Clustering algorithm groups detected anomalous data into clusters based on similarity of user activity feature characteristics. The clustering facilitates administrator categorization of occurring anomaly types along with threat severity levels to the system.Research data were obtained from user activity log records of the clinic information system at Klinik Utama RIDDA Payakumbuh, which underwent preprocessing stages including data cleaning, feature transformation, value normalization, and handling of missing values.Test results demonstrate that the combination of LOF and K-Means achieved accuracy of 89.5%, precision of 87.3%, and recall of 85.7% on the test dataset. These validation metrics prove that the method effectively addresses user behavior deviation detection in the clinic environment. The test results affirm that the hybrid approach can identify suspicious activities with minimal error rates, ensuring reliability. The research contribution provides practical impact for clinic information system administrators in supervising patient data security through integrated early warning mechanisms.

Analisis Komparasi Convolutional Neural Network dan Learning Vector Quantization dalam Klasifikasi Khat Arab Digital

Sabri T Rahman — 2026-06-20

Arabic khat is a form of writing that possesses complex visual characteristics, such as variations in letter shapes, stroke thickness, texture, and stylistic differences. This complexity creates challenges in manually recognizing different types of khat. This study aims to analyze and compare the performance of Convolutional Neural Network (CNN) and Learning Vector Quantization (LVQ) methods in classifying five types of Arabic khat digital images, namely Diwani, Farsi, Naskh, Ruqaa, and Tuluth. The dataset was obtained from the Kaggle.com platform. CNN architecture consists of an input layer of 100×100×1, followed by two convolutional layers with 32 and 64 filters of size 3×3, each followed by ReLU activation and max pooling with stride 2. The network then includes a fully connected layer with 64 neurons, a final fully connected layer corresponding to the number of classes, a softmax layer, and a classification layer. CNN training was conducted using 5-fold cross-validation, applying data augmentation in each fold. For the LVQ method, Local Binary Pattern (LBP) was used for feature extraction from 100×100 images with parameters: radius 1, 8 neighbors, cell size [48 48], and L2 normalization. The extracted features were used for training with an initialization of 25 prototypes from 5 classes. The process also employed 5-fold cross-validation. From 40 testing samples, the CNN model achieved an accuracy of 87.5%, while the LVQ model achieved an accuracy of 85%. The CNN algorithm demonstrated better performance in handling the complex visual patterns of Arabic khat. Meanwhile, LVQ showed advantages in architectural simplicity and computational efficiency. This research is expected to contribute to the development of Arabic khat image classification systems and serve as a reference in selecting optimal methods for Arabic khat recognition.

Impelementasi Sistem Presensi Wisuda Berbasis QR Code untuk Meningkatkan Layanan Menggunakan MERN Stack

Puja Hanifah — 2026-06-30

Graduation is a ceremonial activity routinely conducted at a university, including at Polytechnic Caltex Riau. The process of conferring degrees at Polytechnic Caltex Riau has been experiencing an increase each year. In the registration process, a website has already been used to facilitate the registration, but on the implementation day, it is still done manually. Like the attendance process for graduates still using paper, as well as parents who must sign in when they arrive. This takes a long time in the process, not to mention the loss of attendance records, duplicate signatures, and so on. Because of these issues, a graduation attendance system was built that will use QR codes in the registration process and was developed using MERN Stack technology. The purpose of developing this system is to facilitate the graduation committee in coordinating activities with parents and graduates. The use of MERN Stack technology is very suitable for development for real-time data, and using QR codes speeds up the existing process. From the research results, all features of the system are operational and can be used based on the UAT testing conducted on three users, namely the admin, super admin, and BAAK. The usability testing yielded a score of 63%, indicating that the system is somewhat difficult to use and the interface needs further improvement.

Implementasi Chatbot NgirimWA guna Optimalisasi Respons Pelanggan dan Peningkatan Kualitas Layanan Pelanggan pada UMKM Pakaian Bayi

Fandevi Maitri — 2026-06-21

The rapid growth of the digital ecosystem encourages SMEs to improve customer service quality through digital communication platforms such as WhatsApp Business. However, the limited automation features available on WhatsApp Business still require customer service activities to be handled manually, especially during peak operational hours. This study aims to analyze the implementation of a knowledge-based chatbot using the NgirimWA platform to improve operational efficiency in the TISUKA baby clothing SME. The research applies a qualitative approach through observation, interviews, and literature studies. In addition, Value, Rare, Inimitable, Organization (VRIO), Strengths, Weaknesses, Opportunities, Threats (SWOT), Business Model Canvas (BMC), and Segmentation, Targeting, Positioning (STP) analyses were conducted to understand the business’s internal and external conditions before chatbot implementation. The findings indicate that the use of the NgirimWA chatbot accelerates customer communication, reduces repetitive manual responses, and improves the consistency of product information delivered to customers. The chatbot also helps business owners reduce operational workload, which previously required approximately 3–5 hours daily to respond to customer inquiries manually. Furthermore, chatbot implementation contributes to better customer service quality by providing faster, more informative, and continuously available responses outside operational hours. This study demonstrates that NgirimWA can serve as a relevant digital solution for SMEs to improve customer service efficiency and support business digital transformation.

Evaluasi YOLOv8 untuk Deteksi Kendaraan pada Simulasi Citra Palang Parkir

Syalomita Pasha Sante — 2026-06-21

This study evaluates the use of YOLOv8 for vehicle detection in an image-based parking barrier simulation. The study was conducted because vehicles near a parking entrance are not always captured under ideal visual conditions. Some vehicles may appear far from the barrier point, partially occluded, captured under low-light conditions, or appear together with other vehicles in a single frame. The data were collected from two sources, namely vehicle photos taken using a mobile phone camera and vehicle images obtained from open internet sources. All images were grouped into five testing scenarios: vehicles in front of the barrier area, vehicles far from the barrier point, partially occluded vehicles, low-light conditions, and crowded areas. The testing process was carried out using the YOLOv8n model with a confidence threshold of 0.5. From a total of 131 test images, the model successfully detected vehicles in 103 images, failed to detect vehicles in 28 images, and produced 0 false detections. The average detection accuracy was 77,6%. The best result was obtained in the crowded area scenario, while the lowest result occurred in the low-light condition scenario. These findings show that YOLOv8n can be used as an initial evaluation for vehicle detection in a parking barrier simulation, although further testing with a live camera and physical devices is still needed. This study contributes to the preliminary evaluation of YOLOv8n for vehicle detection in parking barrier image simulation. The main contribution lies in examining the model’s ability to recognize vehicles under several visual conditions, including vehicles in front of the barrier area, vehicles far from the barrier, partially occluded vehicles, low-light conditions, and crowded areas. This study is not intended to represent a complete implementation of an automatic parking barrier system. Instead, it serves as an image-based preliminary evaluation to identify the potential and limitations of YOLOv8n before further development using live cameras and physical parking barrier devices.

Sistem Informasi Monitoring Komoditas Sayur untuk Penentuan Harga Wajar Menggunakan Metode Standar Deviasi

Muh Zia Ulkhaq — 2026-06-21

The dynamic fluctuation of vegetable commodity prices often creates uncertainty in determining fair prices in traditional markets. This condition leads to information asymmetry between traders, buyers, and market managers, requiring a more objective approach to define fair price boundaries. This study aims to develop a statistical-based price monitoring model to identify a more measurable fair price range for vegetable commodities. The method used in this study is standard deviation as a statistical tool to measure the level of price dispersion based on historical price data. The mean and standard deviation values are used to construct the lower and upper bounds of the fair price range using the mean ± 1? approach. The data used in this study were obtained from the red chili commodity at the Banjarnegara Main Market as a case study.The results show that the standard deviation method is able to objectively represent price variability and classify market conditions into three categories, namely low price, normal price, and high price, based on the position of prices relative to the statistical range. In the case study of April 2026, the mean price was Rp63,300/kg and the standard deviation was Rp6,830/kg, resulting in a fair price range between Rp56,470/kg and Rp70,130/kg. This study concludes that the standard deviation approach is effective in identifying fair price boundaries based on historical data. The main contribution of this research is the development of a more objective price analysis model to support price transparency in traditional markets.

Prediksi Saham Berdasarkan Data Teknikal Serta Fundamental Menggunakan Algoritma XGBoost

Yoga Nur Pradana — 2026-06-21

The capital market has an important role in the economy as a means of investment and fundraising, with banking sector stocks being one of the main contributors to market capitalization in Indonesia. However, the investment decision-making process often faces obstacles in the form of limited investors' ability to comprehensively analyze fundamental and technical data, as well as irrational behavior that causes decisions to be less than optimal. This conditions encourage the need for a more objective and data-driven approach to help predict stock price movements. The results of the model evaluation on the test data showed excellent performance: BCA obtained a MAPE of 2.8% and an R² of 0.9488; BNI with MAPE 3.06% and R² 0.8863; Bank Mandiri with a MAPE of 4.70% and R² 0.9114; and BRI with MAPE of 2.48% and R² 0.8872. Based on this model, the results of the share price prediction for 2026 show that BCA is predicted to experience a significant increase from IDR 7,756 (January) to IDR 7,846 (June), while Bank Mandiri is predicted to grow from IDR 5,211 (January) to IDR 5,930 (June). BNI and BRI are predicted to experience an increase in share prices, respectively from IDR 3,327 (January) to IDR 3,683 (June) and from IDR 4,124 (January) to IDR 4,541 (June). This research contributes by presenting a stock prediction model that combines technical and fundamental data at once, applied to four major Indonesian banks Bank Central Asia, Bank Rakyat Indonesia, Bank Mandiri, dan Bank Negara Indonesia in a single modeling framework. This approach has proven to produce good accuracy with an average MAPE of 3.13% and R² 0.919, as well as being a more objective alternative for investors in analyzing stock price movements. However, the prediction results obtained in this study are analytical tools and are not intended as direct investment recommendations.

Perbandingan Grid Search dan Random Search untuk Optimasi Hyperparameter Random Forest pada Klasifikasi Kanker Payudara

Fiona Yenisya Dewi — 2026-06-30

Breast cancer is the most prevalent type of cancer in Indonesia, with 71% of patients diagnosed at advanced stages due to limited access to early detection. This condition necessitates the development of machine learning-based screening systems that are not only accurate but also computationally efficient to enable widespread implementation in healthcare facilities with limited resources, making the selection of an efficient hyperparameter optimization method crucial. This study compares two hyperparameter optimization methods, namely Grid Search and Random Search, applied to the Random Forest algorithm using the UCI Wisconsin Diagnostic Breast Cancer Dataset (569 samples, 30 numerical features) with an identical search space encompassing 288 hyperparameter combinations and stratified 5-fold cross-validation. Experimental results demonstrate that Random Search RF achieves performance equivalent to Baseline RF on threshold-based metrics (accuracy 0.9737; F1-Score 0.9630) while producing the highest AUC-ROC of 0.9950 in 88,26 seconds. In contrast, Grid Search RF yields performance below the baseline (accuracy 0.9561; F1-Score 0.9367) with a computation time of 526,73 seconds, attributable to the optimizer's curse phenomenon in which the selected combination based on cross-validation does not produce optimal generalization on the test data. Random Search is proven to be 5.97 times more efficient than Grid Search with superior solution quality, empirically confirming the theoretical proposition that Random Search is capable of finding competitive configurations at substantially lower computational cost compared to exhaustive search in high-dimensional hyperparameter spaces.

Redesign UI/UX Website Menabung.id Menggunakan Metode User Centered Design dengan Evaluasi UX

Sirojul Munir — 2026-06-30

The use of Menabung.id as a digital platform to record and monitor student savings in schools revealed limitations in both the user interface and user experience. Less intuitive navigation, inconsistent visual presentation, and suboptimal information display created obstacles for teachers and students in utilizing the system effectively. This situation highlighted the urgency of redesigning the UI/UX to make user interactions more comfortable, efficient, and effective, thereby achieving the goals of digitalizing administration and improving financial literacy. This study aimed to redesign the UI/UX of Menabung.id using the User-Centered Design (UCD) approach, placing users at the center of all design stages, including requirement analysis, prototype development, and evaluation. An interactive prototype was developed using Figma, and usability testing was conducted using the System Usability Scale (SUS) and Single Ease Question (SEQ) to measure system effectiveness and ease of use. The results indicated that the UCD-based redesign improved interface quality and user experience. SUS scores reached 81,3 for teachers and 82.5 for students, categorised as Excellent, while SEQ scores were 5.96 and 5.81, indicating a high level of task completion ease. These improvements included teacher and student dashboards, transaction history, savings schedules, real-time notifications, and an interactive chatbot. The findings demonstrated that user-centered design effectively enhanced usability and user satisfaction. Further research can complement the evaluation with eye-tracking or A/B testing methods and develop the system for mobile platforms to increase user accessibility.

Integrasi Principal Component Analysis dan Logistic Regression untuk Analisis Sentimen Kepuasan Pelanggan Berdasarkan Ulasan Online

Tsalsabila Jilhan Haura — 2026-06-21

Customer reviews on digital platforms are an important source of information for evaluating service quality and customer satisfaction levels. However, the unstructured nature of review data and its high feature dimensionality pose challenges in the sentiment analysis process. This study aims to develop a customer sentiment analysis model by integrating Principal Component Analysis (PCA) and Logistic Regression. The data used are 679 Indonesian-language reviews obtained through web scraping techniques from Google Reviews at ten d'Besto EBM branches in Padang City. The research stages include text preprocessing, TF-IDF weighting, dimensionality reduction using PCA, and sentiment classification using Logistic Regression. The results show that PCA is able to reduce data complexity by producing two principal components that explain 85.7% of the total data variance. The Logistic Regression model built on the features resulting from PCA reduction achieved an accuracy of 82%, demonstrating the model's ability to effectively classify positive and negative sentiments. In addition to improving computational efficiency, the use of PCA also helps reduce feature redundancy in high-dimensional text data. The contribution of this research is to produce a simpler and more efficient sentiment analysis approach to process customer reviews and provide data-based information that can be used to support service quality evaluation and decision-making in the culinary industry.

Optimasi Klasifikasi Hate Speech dan Offensive Language melalui Frozen RoBERTa Feature Extraction dan Random Forest

Marsha Cahyani Dwisyakilla — 2026-06-24

Hate speech and offensive content detection on social media remains a significant challenge in Natural Language Processing (NLP) due to the characteristics of Twitter data, which are typically short, informal, and contain various elements such as mentions, URLs, hashtags, and emotional expressions that complicate the classification process. End-to-end Transformer fine-tuning approaches generally require substantial computational resources; therefore, this study explores a more computationally efficient approach by utilizing RoBERTa as a frozen feature extractor combined with Random Forest as the classifier. This approach enables the exploitation of contextual representations generated by Transformer models without requiring full model retraining.The study employs the HASOC 2021 English Track dataset, which consists of two classification tasks: Task A for binary classification (HOF and NOT) and Task B for multi-class classification (HATE, OFFN, PRFN, and NONE). The classification pipeline is optimized through the incorporation of handcrafted features, oversampling, Random Forest hyperparameter tuning, and threshold tuning in specific scenarios. Model performance is evaluated using accuracy, precision, recall, and F1-macro, with F1-macro serving as the primary metric due to class imbalance. The best-performing model achieved an F1-macro score of 0.80 on Task A and 0.64 on Task B. These results indicate that the combination of frozen RoBERTa representations and Random Forest provides strong performance for binary hate speech and offensive content classification. However, the performance on Task B highlights the difficulty of distinguishing linguistically similar categories, such as HATE, OFFN, and PRFN, suggesting that fine-grained multi-class classification remains a challenging task. Overall, the findings indicate that RoBERTa-based frozen feature extraction constitutes a computationally efficient alternative for hate speech detection on English Twitter data, although further improvements are required to enhance performance in multi-class classification settings.

SIMPATA: Integrasi Administrasi dan Monitoring Progres Kegiatan Magang Berbasis Web

Afif Rifai Nimal Abdu — 2026-06-24

Internship management at the Statistics Indonesia Office of Sukoharjo Regency still relies on several separate and unintegrated media, covering document submission and verification, attendance recording, activity documentation, and participant task-progress monitoring. This condition causes data to be scattered, makes the recapitulation process less practical, and prevents participant activities from being monitored through a single system. This study aims to design and develop SIMPATA (Internship Participant Management and Governance Information System) as a web-based system that integrates internship administration and activity monitoring. The system was developed using an adapted Extreme Programming approach through the stages of planning, design, coding, and testing. Research data were collected through interviews, observations, and literature studies. The system was designed using the Unified Modeling Language, while its implementation employed PHP, the Laravel framework, and a MySQL database. The main contribution of SIMPATA lies in integrating registration, document verification, acceptance-status determination, participant and supervisor management, digital attendance, activity logbooks, task assignment, work submission, and progress monitoring into a single system. Black-box testing involving ten test scenarios showed that all major functions operated according to the specified functional requirements. Furthermore, User Acceptance Testing involving one administrator, three supervisors, and ten students obtained an overall score of 88.71%, which was classified in the Strongly Agree category. These results indicate that SIMPATA was well accepted by its users and can support more centralized and well-documented internship administration and activity monitoring.

Analisis Limitasi Performa Penilaian Esai Otomatis pada Aplikasi ESAO Berdasarkan Metrik BLEU dan ROUGE

Akhmam Fahmi — 2026-06-24

The development of GenAI has encouraged the use of automated essay scoring technology through various platforms, one of which is the ESAO (Essay Analytic Online) application. Although this LLM-based system is capable of automatically generating assessment feedback narratives, standardizing evaluation methods to measure the reliability of these texts still faces significant challenges. This study aims to test the suitability of the Bilingual Evaluation Understudy (BLEU) and Recall-Oriented Understudy for Gisting Evaluation (ROUGE) metrics as instruments to measure the extratextual performance of the ESAO application. The research method was carried out by comparing feedback texts from ESAO with authentic lecturer assessment drafts on three different characteristics of the exam material: dataset condition analysis, descriptive statistics, and correlation and regression. The test results showed an average value of the BLEU metric of 0.0522 and ROUGE of 0.1255. This study revealed that low scores do not represent a functional failure of the ESAO application, but rather indicate fundamental limitations and shortcomings in using rigid lexical metrics (word-based metrics) in assessing dynamic generative texts. The BLEU and ROUGE metrics rely heavily on rigid n-gram overlap, thus failing to capture the semantic similarity, academic reasoning context, and linguistic variation generated by ESAO. This study concludes that traditional evaluation metrics such as BLEU and ROUGE are inaccurate and incompatible as a single benchmark for Generative AI performance in the context of educational assessment, necessitating a transition to semantic-based metrics in the future.

Pengembangan Sistem Informasi Akademik Terintegrasi Berbasis Web Mobile Pada Lembaga Pendidikan Menengah Atas

Sandy Maulana Rifqi — 2026-06-24

Academic information dissemination at Senior High School has been carried out conventionally, often resulting in delays in providing attendance, class schedules, and grade reports to students and parents. The limitation of non-real-time information access impacts the lack of parental supervision regarding student discipline and academic progress. This research aims to design and build a web and mobile-based Academic Information System capable of integrating school data management digitally, quickly, and accurately. The system development method used is the Waterfall method, including requirement analysis, system design, implementation, testing, and maintenance. The system was developed using the Laravel framework for the web-based admin dashboard and the Flutter framework for the mobile application for students and parents, supported by a MySql database. This research provides a scientific contribution in the form of a cross-platform integration architecture (dual-framework) connected via RESTful API to overcome data communication asymmetry in educational institutions. This cross-platform integration is implemented to facilitate administrators' needs in large master data management while providing enhanced accessibility for mobile users. Functional testing using the Black Box Testing method showed that all main features operate as expected, and User Acceptance Testing (UAT) resulted in an average score of 86.0%, falling into the "Very Good" category. The results of this study indicate that this integrated information system is effective in facilitating access to academic information transparently and improving administrative efficiency in educational institutions.

Deteksi Kondisi Terumbu Karang Menggunakan YOLO versi 8 pada Citra Bawah Laut Secara Real-Time

Keysia Lestari Sasikome — 2026-06-25

Coral reef ecosystems play an important role in maintaining the balance of the marine environment and supporting the marine tourism sector. However, coral reef damage due to climate change, pollution, and human activities continues to increase, requiring efficient and sustainable monitoring methods. This study aims to develop a coral reef condition detection system based on the YOLOv8 method by utilizing real-time underwater imagery. The research dataset was obtained from the coral reef conservation area of ??Bahoi Village, West Likupang, North Sulawesi. The research stages include dataset collection, image preprocessing, data augmentation, object annotation, YOLOv8 model training, model performance evaluation, and web-based detection system implementation. Model evaluation was carried out using precision, recall, mean Average Precision (mAP), and confusion matrix metrics. The test results showed that the YOLOv8 model was able to detect coral reef objects with good performance, indicated by a precision value of 76.51%, recall of 98.57%, mAP50 of 86.78%, and mAP50-95 of 86.77%. Confusion matrix analysis showed that the model did not misclassify coral reef species, while a small number of errors occurred only in objects detected as background due to underwater environmental conditions such as water turbidity and light refraction. The results showed that YOLOv8 is effective for detecting and monitoring coral reef conditions automatically and in real time, thus potentially supporting conservation activities and sustainable marine ecosystem management.

Penerapan Extreme Programming dalam Pengembangan Sistem E-Tiket Berbasis Web pada Objek Wisata Lokal

Tarso Tarso — 2026-06-25

A local tourism object serving as the research site still relies on a manual ticketing system, which has caused several operational issues including long queues at the entrance, transaction recording errors, and limited access to real-time visitor data that hinders management decision-making. This study aims to develop a web-based e-ticketing system as a solution to these problems by implementing the Extreme Programming (XP) method. XP was selected because it supports iterative and adaptive software development in response to changing user requirements through four main phases: planning, design, coding, and testing. The system was developed using PHP Native as the server-side programming language, MySQL as the database management system, and HTML, CSS, and JavaScript for the user interface, along with a ticket booking feature and WhatsApp-based notification. Testing was conducted using Black Box Testing to verify system functionality and User Acceptance Testing (UAT) to measure the level of end-user acceptance. Black Box Testing results on five main system functions showed that all features performed according to functional requirements with a 100% success rate. UAT results involving 15 respondents consisting of tourism managers and visitors yielded an average feasibility score of 88.03%, categorized as "Very Feasible." This study demonstrates that the XP method is effective in producing a functional web-based e-ticketing system that is accepted by end users. This study contributes by demonstrating the effectiveness of the Extreme Programming (XP) method in the context of local nature-based tourism, a domain that has received limited attention in previous research. In addition, the study provides a more comprehensive system evaluation through the combined use of Black Box Testing (100% success rate) and User Acceptance Testing (UAT) with a score of 88.03%, resulting in a more thorough validation compared to previous studies employing the XP methodology.

Prediksi Minat Pencarian Layanan Pesan-Antar Makanan Online (GoFood dan GrabFood) di Indonesia Menggunakan Algoritma Random Forest Regression dengan Walk-Forward Validation

Sinta Bella — 2026-06-25

The rapid growth of online food delivery services in Indonesia, particularly GoFood and GrabFood, creates significant operational challenges due to unpredictable fluctuations in user interest that cause driver and merchant capacity imbalances. Actual transaction data is proprietary, necessitating a proxy-data approach using Google Trends search interest indices. This study predicts GoFood search interest in Indonesia using Random Forest Regression based on Google Trends data from January 2018 to December 2025 (96 monthly records). The primary contributions of this study are threefold: first, the application of walk-forward validation as a methodologically sound evaluation approach for time-series data that eliminates temporal data leakage; second, the use of lag features (GoFood_lag1 and GrabFood_lag1) ensuring all predictor variables are practically available at prediction time; and third, empirical validation that this approach yields more conservative and scientifically defensible evaluations compared to conventional random split methods. Evaluation results yield MSE 1.4510, RMSE 1.2046, R² 0.5065, and MAPE 6.55%, demonstrating adequate generalization capability for data-driven operational planning.

Pengembangan Chatbot Customer Service dengan Retrieval-Augmented Generation pada Usaha Mikro Kecil Menengah KampusMadu

Ahmad Zakky Zamani — 2026-06-30

Customer service in Micro, Small, and Medium Enterprises (MSMEs) is generally still conducted manually, resulting in delayed responses, inconsistent information, and high operational workload. This study aims to develop and evaluate a customer service chatbot based on Large Language Model using the Retrieval-Augmented Generation (RAG) method as an automation solution for customer service at KampusMadu MSME. The system was developed using an adaptive Waterfall approach encompassing requirements analysis, design, implementation, and testing stages. Testing was conducted through three approaches: functional testing using Black Box Testing method, retrieval accuracy testing using Precision@3, Recall@3, and Hit Rate@3 metrics, and usability testing using the System Usability Scale (SUS) involving 10 respondents. The test results show that all system functions operate as required, with average Precision@3 of 63.3%, Recall@3 of 72.5%, and Hit Rate@3 of 100%. The SUS score obtained was 69.75 (category "OK"), indicating that the system is reasonably acceptable to users. The chatbot system was successfully integrated into the KampusMadu website as an interactive widget directly accessible to customers. This study demonstrates that the application of RAG in MSME chatbots can effectively support the automation of customer service in a more responsive, informative, and contextual manner.

Rancang Bangun Sistem Keamanan Laser Berbasis Internet of Things Menggunakan Metode Prototyping dengan Peringatan Dini dan Bukti Visual Real-Time Untuk Mencegah Pencurian

Randikal Hikrenc Menono — 2026-06-30

The rapid development of digital technology, particularly in the field of the Internet of Things (IoT), has brought significant changes to various aspects of human life, including security systems. Security is a crucial aspect for business owners, especially those operating in environments with a high risk of theft. This study aims to design and develop an IoT-based laser security prototype capable of providing real-time early warning notifications to users. The research employed the Prototyping method, which consists of communication, quick plan, modeling quick design, construction of prototype, and deployment and feedback stages. The system utilizes an ESP32-S3 Cam microcontroller as the main controller integrated with an LDR sensor and a SIM800L GSM module. The system operates by detecting interruptions in a laser beam caused by objects passing through the monitored area, which serve as intrusion triggers. When the laser beam is interrupted, the light intensity received by the LDR decreases significantly, prompting the microcontroller to automatically instruct the SIM800L module to place a phone call to the user. In addition, the ESP32-S3 Cam captures images of the monitored area and sends them through a Telegram bot as visual evidence. The testing results indicate that the system performs effectively, with an average response time of 8 seconds from intrusion detection to phone call notification. The success rate of phone call notifications reached 90%, while the visual evidence transmission through Telegram achieved a success rate of 100%. These results demonstrate that the system is capable of providing real-time alerts and visual information to users. The contribution of this research lies in the successful integration of intruder detection based on laser beam interruption, alarm notifications through telephone calls using the SIM800L module, and visual evidence transmission using the ESP32-S3 Cam and Telegram application into a single Internet of Things (IoT)-based security system capable of providing early warning notifications. When a laser beam interruption is detected, the system can automatically notify users through a phone call and send visual evidence in real time, thereby enhancing the effectiveness of remote security monitoring and response. However, the system performance is still affected by the stability of the GSM network used by the SIM800L module and is limited to detecting objects that interrupt the laser beam path. Therefore, the proposed system offers a smart, responsive, and cost-effective security solution that enables remote monitoring and supports theft prevention.

Pengembangan Aplikasi berbasis Virtual Reality untuk Sistem Teaching Robot pada SmartEdu Station

Yuliadi Erdani — 2026-06-30

Conventionally, the trajectory programming procedure for industrial robots using physical teach pendants imposes a high cognitive load, is prone to triggering structural collision accidents for novice students, and limits training efficiency due to the minimal availability of equipment. This study aims to solve these issues by developing an autonomous direct teaching system for the UR5e manipulator utilizing Virtual Reality (VR) technology and a Digital Twin architecture at the SmartEdu-Station facility. As a solution, the system captures global spatial coordinate inputs from the user's natural hand movements via 6-DoF VR controller tracking, then translates them into real-time virtual robot posture visualization using Inverse Kinematics computation. The trajectory coordinate sets are validated for safety as autonomous waypoint data before being transmitted to the physical controller. Empirical testing results prove that data communication via TCP/IP and RTDE protocols records a trajectory packet transmission success rate of 100%, with an average execution response delay consistently under 1 second. Furthermore, interface ergonomics testing using the System Usability Scale (SUS) questionnaire on 10 respondents yielded an average score of 79.0 (Good Category). This research contributes significantly by providing an offline teaching simulation framework that successfully eliminates the risk of physical equipment damage while interactively and safely reducing the learning curve of spatial mapping.

Penerapan XGBoost dan SMOTE untuk Klasifikasi Metode Pembayaran Pelanggan pada Data Transaksi Tidak Seimbang

Fadilah Nuria Handayani — 2026-06-30

The increasing use of digital payment methods in retail transactions highlights the importance of analyzing customer payment behavior. This study aims to classify customer payment methods using the XGBoost algorithm and to evaluate the effect of Synthetic Minority Over-sampling Technique (SMOTE) in handling class imbalance. The dataset consists of 287,422 transaction records processed using the Cross Industry Standard Process for Data Mining (CRISP-DM) framework, which includes data understanding, data preparation, modeling, and evaluation stages. Experimental results show that the XGBoost model without SMOTE achieved an accuracy of 92.83% and a ROC-AUC of 0.7759, but performed poorly in identifying the minority class (Card), with a recall of 0.14, indicating a strong bias toward the majority class. After applying SMOTE, the model’s ability to detect the minority class improved, with recall increasing to 0.53 and F1-score reaching 0.28, although accuracy decreased to 78.50% and ROC-AUC to 0.7529. This study contributes by implementing XGBoost combined with the SMOTE method for customer payment method classification on imbalanced data and evaluating model performance using multiple classification metrics. This trade-off indicates that SMOTE improves sensitivity toward minority classes while affecting overall predictive accuracy. The findings highlight that evaluation of imbalanced classification models should not rely solely on accuracy but must also consider precision, recall, F1-score, and ROC-AUC to obtain a more comprehensive assessment. Overall, while SMOTE enhances minority class detection, further improvements are still required to achieve more stable and reliable classification performance.

Pemodelan Topik pada Komentar Media Sosial X menggunakan Latent Dirichlet Allocation

Ardelia Adzra — 2026-06-30

Sexual harassment is a social issue widely discussed on the social media platform X. However, the high volume of unstructured comments makes it difficult to manually identify the main topics of discussion. This study aims to identify the main topics in comments related to sexual harassment on X using the Latent Dirichlet Allocation (LDA) method. The data used consist of comments on the topic of sexual harassment collected from X during the 2024–2026 period. The research stages include data collection, data preparation, dictionary and corpus construction, LDA modeling with hyperparameter tuning, evaluation using coherence score, and topic interpretation based on dominant keywords and representative data. The results show that the best LDA model consists of four topics with a coherence score of 0.517. These four topics are interpreted as Handling Cases of Sexual Harassment in Educational Environments, Victims’ Experiences and Psychological Impacts, Cases of Sexual Harassment in Higher Education, and Protection Related to Sexual Harassment. These findings indicate that the LDA method is capable of identifying the main topics in sexual harassment comments and helping to organize unstructured social media data into information that is easier to understand. The contribution of this study is the proposed Latent Dirichlet Allocation (LDA)-based topic modeling approach with hyperparameter tuning to identify and organize unstructured sexual harassment comments on the social media platform X into coherent and interpretable topic clusters. The resulting topic mapping provides valuable insights into the issues that receive the greatest public attention and can serve as a foundation for understanding public concerns. Furthermore, these findings have the potential to support the development of victim support services, including telemedicine-based systems.

Implementation of the Analytical Hierarchy Process (AHP) with AI-Assisted Validation for Waste Processing Method Selection

Agung Firdausi Ahsan — 2026-06-30

The increasing volume of municipal solid waste in Sumenep Regency has created significant challenges for local authorities in selecting an effective and sustainable waste treatment method. The selection process requires consideration of multiple criteria, including economic, technical, environmental, social, labor, and material recovery aspects. Therefore, this study aims to determine the most suitable waste processing method by applying the Analytical Hierarchy Process (AHP) as the primary decision-making approach and an AI-assisted validation approach as a comparative evaluation tool. The AHP method was used to calculate the relative importance of criteria and rank three waste treatment alternatives, namely Composting, Sanitary Landfill, and Incineration, based on expert judgments. To strengthen the reliability of the decision-making process, an AI-assisted evaluation using a Large Language Model (LLM) was conducted to assess the same alternatives according to the established criteria and compare the resulting rankings with those obtained from AHP. The results of the AHP analysis indicate that Composting has the highest priority weight of 48.0%, followed by Sanitary Landfill with 33.5% and Incineration with 18.5%. Similarly, the AI-assisted evaluation generated the highest score for Composting (0.9835), followed by Sanitary Landfill (0.6130) and Incineration (0.6025). The consistency between the rankings produced by AHP and the AI-assisted assessment demonstrates the robustness of the selected alternative. The findings suggest that Composting is the most appropriate waste treatment method for Sumenep Regency due to its superior environmental performance, social acceptance, and material recovery potential. Furthermore, the study highlights the potential of AI-assisted evaluation as a supporting validation tool for enhancing multi-criteria decision-making in waste management planning.

Implementasi Data Mining K-Means Clustering Untuk Pengelompokan Produk Keramik Berdasarkan Frekuensi, Volume, dan Jangkauan Penjualan

Ferdian Arya Dinata — 2026-06-30

Ceramic inventory management at CV. Makmur Bersama has generally relied on intuition or partial sales data, without accounting for purchasing behavior patterns as a whole. This approach simultaneously creates two major risks: overstocking of slow-moving products, which burdens working capital and storage space, and stockouts of high-demand products, which can result in lost sales opportunities. This problem is further compounded by the limitation of stock data, which typically contains only a single quantitative variable such as the number of units sold and is therefore unable to comprehensively capture product demand characteristics, such as how frequently a product is purchased or how broad its customer base is. As a result, restocking decisions and promotional strategies are often poorly targeted. This research applies the K-Means algorithm to cluster ceramic products based on historical sales patterns as a solution to this limitation. Historical sales data from CV. Makmur Bersama for the 2025 period, consisting of 6,328 transactions, was processed into 417 unique products through a feature engineering approach using Frequency, Monetary, and Reach (FMR) namely transaction count, total quantity sold, and unique customer count per product. After outlier detection using the Interquartile Range (IQR) method, 381 products remained for the clustering process. The optimal number of clusters was determined using the Elbow Method, resulting in k=4 as the best cluster count. Evaluation using the Davies-Bouldin Index (DBI) produced a value of 0.8954, categorized as good, and stability testing across five iterations with different random states showed consistent results (DBI standard deviation of 0.0034). The clustering results produced Cluster 1 (190 products, 49.9%) as slow-moving products, Cluster 2 (34 products, 8.9%) as top-performing products with an average transaction frequency of 30.8 times, Cluster 3 (93 products, 24.4%) as potential products, and Cluster 4 (64 products, 16.8%) as products with limited demand. This research provides practical contributions for companies in determining restocking priorities, promotional strategies, and working capital efficiency based on actual sales patterns. This research contributes methodologically through the adaptation of the RFM framework into FMR to better suit real-world data constraints, as well as the integration of the Elbow Method, Davies-Bouldin Index, and stability testing as a comprehensive validation mechanism. Practically, the segmentation results can be directly utilized by the company as a basis for restocking priorities, promotional strategies, and working capital allocation efficiency based on actual sales patterns.

Klasifikasi Persepsi Publik Terhadap Perang Dagang Amerika Serikat Menggunakan Algoritma Naïve Bayes Classifier

Bunga Nurul Manisa — 2026-06-30

The import tariff policy implemented by the President of the United States on April 2, 2025 triggered tensions in global trade and provoked various public reactions. Differences in public perceptions of the policy generated diverse opinions, including support, criticism, and neutral responses, making sentiment analysis necessary to understand public opinion trends more systematically. This study aims to classify public perceptions of the U.S. trade war through sentiment analysis of Twitter data using the Naïve Bayes Classifier (NBC) algorithm. The dataset consists of 2,000 tweets collected using the keywords “trade war” and “import tariff increase” during April 3–30, 2025. Six preprocessing stages were applied: cleaning, case folding, tokenizing, slangword normalization, stopword removal, and stemming to improve data quality and consistency. Automatic labeling was conducted using a lexicon-based method with the InSet dictionary, yielding sentiment distributions of 83.5% negative, 12.8% positive, and 3.8% neutral. Feature representation was performed using TF-IDF, followed by an 80:20 train-test split. To address class imbalance, the Synthetic Minority Over-sampling Technique (SMOTE) was applied. Experimental results show that the NBC model without SMOTE achieved an accuracy of 83.5% but exhibited bias toward the majority class. After applying SMOTE, the dataset became balanced with 1,335 samples per class. Although overall accuracy decreased to 76%, the Macro F1-Score improved from 0.30 to 0.45, indicating improved model performance in handling multi-class classification more fairly. Additionally, the model achieved a recall of 43% for the positive class and 13% for the neutral class, providing a more representative evaluation of public sentiment toward the U.S. trade war issue.

Perancangan Keamanan Informasi pada Sistem Persuratan Berbasis Web Menggunakan Autentikasi dan Middleware

Panggah Widiandana — 2026-06-30

The management of digital correspondence through web-based systems has improved administrative efficiency, but on the other hand, it introduces new vulnerabilities to cyber threats. Information security is a crucial aspect in the development of web-based mailing systems because this system manages incoming mail, outgoing mail, and official documents that are highly sensitive. Without a good security mechanism, the system has the potential to experience unauthorized access, data manipulation, and information leaks. Therefore, this study aims to design an information security mechanism through the application of authentication (auth) and middleware as system access controllers. The research method used is a system design method that includes security requirement analysis, Access Control Matrix modeling, routing interceptor architecture design, and Black-box authorization testing. In its application, authentication is used to ensure user identity before accessing the system, while middleware serves as a security layer to restrict access based on user roles. The design results show that the application of auth and middleware can improve access control, maintain data confidentiality, and strengthen information security in web-based mailing systems. The main contribution of this research is the provision of an integrated Role-Based Access Control (RBAC) layered information security model that can be adopted by various institutions to prevent privilege escalation and data manipulation in internal administration systems.

Implementasi Data Mining untuk Menentukan Pola Pembelian Obat Menggunakan Metode Apriori

Muhtajuddin Danny — 2026-06-30

The development of information technology has increased the amount of drug sales transaction data in the pharmacy sector. However, transaction data are generally used only as administrative archives and have not been optimally utilized to produce strategic information. This study aims to implement data mining using the Apriori method to determine drug purchasing patterns based on pharmaceutical transaction data. This research employed a quantitative approach using the Pharmacy Transactional Dataset obtained from the Kaggle platform. The research stages were conducted using the Cross Industry Standard Process for Data Mining (CRISP-DM), including business understanding, data understanding, data preparation, modeling, evaluation, and deployment. The analysis process was carried out using the Python programming language with the assistance of the pandas and mlxtend libraries. The results showed that the purchasing relationship between Paracetamol and Vitamin C had the highest association value with a support value of 32% and a confidence value of 78%. These results indicate that the Apriori algorithm is capable of identifying relationships among drug products based on pharmaceutical transaction data. The resulting information can be utilized to support promotional strategies, drug inventory management, and business decision-making in the pharmaceutical sector.

Security Assessment of E-Commerce Website Using NIST SP 800-115 Based on OWASP Top 10

Arif Setyo Wibowo — 2026-06-30

The rapid growth of e-commerce platforms in Indonesia has increased the risk of cyber threats targeting sensitive user data, including personal information and payment details. PT. XYZ, a mattress company that recently launched its first e-commerce website, has attracted 42,222 visitors and generated revenue of Rp994,878,300 within its first six months, yet has never undergone any form of security testing. This raises serious concerns, as undetected vulnerabilities may expose the platform to identity theft, data breaches, and unauthorized access. This study aims to identify existing security vulnerabilities, determine the severity level of each finding, and provide concrete remediation recommendations before those vulnerabilities are exploited. The assessment was conducted using the NIST SP 800-115 framework across four phases: Planning, Discovery, Attack, and Reporting, with vulnerability classification based on OWASP Top 10 (2021). The Discovery phase utilized Google Dorking, WHOIS, wfuzz, Wappalyzer, Nmap, Burp Suite, and OWASP ZAP to gather intelligence and identify weaknesses. The Attack phase successfully exploited six confirmed vulnerabilities: Clickjacking, CSP Header Not Set, Vulnerable JS Library, Cross-Domain Misconfiguration, Source Code Disclosure, and Username Enumeration and Brute Force, mapped to OWASP categories A05, A06, and A07, with risk levels ranging from Medium to High. This research contributes by demonstrating that newly deployed platforms are not inherently secure and that integrating NIST SP 800-115 with OWASP Top 10 provides a structured approach to identifying real security vulnerabilities in e-commerce systems.

Klasifikasi Tingkat Kepuasan Pengguna Produk Body Care Menggunakan Algoritma Decision Tree

Nur Jannah Hasibuan — 2026-06-30

The increasing competition in the body care industry encourages companies to understand customer satisfaction as a basis for improving product quality and service performance. However, analyzing user satisfaction often produces complex data that are difficult to process manually. This study aims to apply the Decision Tree algorithm to classify the satisfaction levels of body care product users based on user characteristics and product evaluations. The research data were collected through questionnaires distributed to 250 respondents, including attributes such as gender, age, frequency of use, product quality, price, service quality, and satisfaction level as the target variable. The research stages consisted of data preprocessing, attribute selection, data transformation, splitting data into training and testing datasets, and building a classification model using the Decision Tree algorithm. Model evaluation was carried out using a confusion matrix with accuracy, precision, recall, and F1-score metrics. The results showed that the model was able to classify user satisfaction into four categories: very satisfied, satisfied, fairly satisfied, and dissatisfied, with an accuracy of 58%, precision of 57%, recall of 57%, and F1-score of 57%. This study contributes to the implementation of data mining for customer satisfaction analysis in the body care industry and helps companies identify dominant factors influencing user satisfaction, particularly product quality and service quality. In addition, the findings are expected to serve as a reference for developing customer satisfaction analysis systems based on data mining in the beauty and body care industry.

Prediksi Konsentrasi CO(GT) Menggunakan Long Short-Term Memory pada Data Sensor Kualitas Udara IoT

Asep Arwan Sulaeman — 2026-06-30

Air quality deterioration has become a major challenge for public health and environmental management in urban areas. Internet of Things (IoT)-based monitoring systems continuously generate sensor data that can be exploited for air quality prediction; however, these datasets commonly contain missing values, noise, and temporal dependencies that may reduce prediction accuracy. This study proposes a Long Short-Term Memory (LSTM)-based model to predict carbon monoxide (CO(GT)) concentrations using the Air Quality UCI dataset, which consists of 9,357 observations and 15 attributes. During preprocessing, -200 values were identified as missing-value indicators, followed by invalid-data handling, Min-Max normalization, and sequence generation using a sliding-window approach with a window size of four. The processed data were divided into training and testing sets using an 80:20 ratio. The prediction model employs a single LSTM layer with 50 hidden units and a Dense output layer and is trained using the Adam optimizer for 50 epochs. Experimental results achieved a Mean Absolute Error (MAE) of 0.0389 and a Root Mean Squared Error (RMSE) of 0.0567, indicating that the proposed model effectively captures temporal patterns in air quality observations with relatively low prediction errors. These findings are consistent with previous studies reporting the effectiveness of LSTM for air quality forecasting and demonstrate its potential to support continuous IoT-based environmental monitoring systems. Future work may incorporate hyperparameter optimization and comparative evaluations with alternative deep learning architectures to further improve predictive performance.

Perancangan UI/UX Fitur Pariwisata pada Aplikasi Garut Hebat Super App Menggunakan Metode Prototipe

Futri Gina Firnanda — 2026-06-30

The tourism feature on the Garut Hebat Super App currently has limitations in navigation structure, presentation of destination information, and a lack of interactive elements, making it difficult for users to obtain tourism information quickly and in a structured manner. This study aims to design the UI/UX of the tourism feature using the Prototype method to produce an interface design that is easier to use and suits user needs. The Prototype method is implemented through the stages of communication, quick plan, modeling quick design, construction of prototype, and delivery and feedback carried out iteratively based on user feedback. The results of the study are interactive prototypes (high-fidelity) that include destination video features, ticket price information, and integrated tourist route maps. Usability evaluation was conducted using the System Usability Scale method on 50 respondents Test results show an average score of 80.09, placing it in the "acceptable" category, with a good level of usability that is acceptable to users. This value indicates that the prototype has a fairly good level of usability and is acceptable to users, although there are still several interface aspects that need to be improved. This research contributes to the design of a prototype-based UI/UX for tourism features on an integrated public service platform (super app) that combines destination information, ticket prices, travel videos, and route navigation in a single interface. The research findings are expected to serve as a reference for developers of regional public service applications in improving the quality of digital tourism services that are oriented towards user needs.

Rancang Bangun Website Billing Pada Penyewaan Playstation Menggunakan Metode Waterfall

Erwin Irawan — 2026-06-30

The process of recording usage duration and calculating tariffs at the AS2 Lalung PlayStation rental business is still carried out conventionally using notebooks and calculators. This method leads to various problems, such as errors in recording the start and end times of play, inaccuracies in rental cost calculations, and difficulties in compiling daily, weekly, and monthly income reports. Furthermore, the manual system cannot properly document transaction history in a structured manner, making it difficult for business owners to monitor revenue and make business decisions. This study aims to design and build a web-based billing system capable of automatically recording rental duration, accurately calculating costs, and presenting structured and real-time income reports. The development method used is Waterfall, which includes the stages of requirement identification, system design (using UML, ERD, and mockups), implementation (using PHP, the CodeIgniter 4 framework, and MySQL database), and testing using a black-box testing approach. The main contributions of this research include the application of a 15-minute time rounding algorithm using the ceil() function, which is more proportional and fair for customers compared to conventional hourly systems, as well as the development of a lightweight, easy-to-implement web-based billing system specifically designed for the internal needs of small-to-medium-scale PlayStation rentals. In addition, this system integrates dynamic PlayStation unit management features and multi-filter income reports that are not yet available simultaneously in similar PlayStation rental systems. The results show that the system successfully records start and end times in real-time, calculates fees based on hourly rates (IDR 5,000 for PS3 and IDR 8,000 for PS4), stores all transaction data in the database, and automatically generates daily, weekly, and monthly income reports. Functional testing on 18 scenarios shows that all key features, such as admin login, PlayStation unit management, rental processing, invoice generation, and report presentation, function as required (100% success rate). With this system, the risk of recording errors can be significantly reduced, and service efficiency increases by more than 75% compared to manual methods. Future research can develop online booking features, digital payment integration, and report export to PDF or Excel formats to further expand system functionality.

Implementasi Sistem Informasi Pengaduan Siswa Berbasis Web dengan Pemantauan Service Level Agreement

Zakhi Febriyan — 2026-06-30

A student complaint system serves as a school service channel allowing students to submit grievances, criticisms, and reports regarding activities and services within the school environment. Complaint management at SMK IPTEK Weru, Sukoharjo Regency, currently lacks digital system support; consequently, report logging is disorganized, resolution processes are time-consuming, and follow-up monitoring is suboptimal. This research aims to design and develop a web-based Student Complaint Information System to provide a more organized and integrated solution for complaint management. The system was built using the Waterfall approach, implemented in stages ranging from requirements identification, design, and application development to functional testing and system maintenance. Research data derived from direct observation, interviews, and literature reviews served as the foundation for system development. The system utilizes Laravel as the application framework and MySQL for data storage. Black-box testing was conducted to ensure that every function operates according to requirements. A key contribution of this research is the implementation of a Service Level Agreement (SLA) feature within the web-based system, enabling the school to monitor the timeliness of complaint resolution in a more structured manner. Results indicate that core features such as login, complaint submission, data management, status monitoring, and report generation function as designed. Meanwhile, User Acceptance Testing (UAT) yielded an acceptance score of 85.00%, placing it in the "Very Good" category. These results demonstrate that the developed system enhances the effectiveness of the complaint process, accelerates report handling, and facilitates complaint management for students, staff, and administrators at SMK IPTEK Weru, Sukoharjo Regency.

Latency and Reliability Evaluation of an HTTP-Controlled ESP8266 Wi-Fi Robot Car Using MIT App Inventor

Ryan Fikri — 2026-06-30

Low-cost Wi-Fi robot cars are frequently presented as functional prototypes, but many studies do not quantify whether motion commands remain reliable and responsive as wireless distance increases. This study therefore designed and experimentally evaluated an ESP8266-based differential-drive robot controlled by an Android application developed with MIT App Inventor. The proposed solution applies an HTTP command-mapping method in which each mobile-interface event is converted into a request, parsed by an ESP8266 web server, mapped to an L298N H-bridge state, and acknowledged after the control action is issued. The objective was to determine functional command accuracy and characterize the relationship among control distance, command-response latency, received signal strength indicator (RSSI), and communication reliability. Five commands forward, backward, left, right, and stop were tested in 150 functional trials, while 500 communication trials were conducted at 1, 5, 10, 15, and 20 m under indoor line-of-sight conditions. The robot correctly executed 148 of 150 functional commands, corresponding to 98.67% success. Distance-based reliability remained 100% at 1–5 m, decreased to 99% at 10 m and 97% at 15 m, and reached 92% at 20 m. Mean command-response latency increased from 88 to 248 ms while RSSI declined from ?38 to ?77 dBm. The main contribution is a reproducible, low-cost evaluation framework that links interface commands, HTTP communication, wireless quality, and physical motion execution. The results indicate that the platform is appropriate for responsive laboratory teleoperation within 15 m, while operation near 20 m requires stronger fail-safe and acknowledgement mechanisms.

Analisis Efektivitas IndoBERT untuk Klasifikasi Multilabel Terjemahan Hadis Bukhari Menggunakan Logistic Regression

Achmad Yamin Harahap — 2026-06-30

Hadith serves as the second source of guidance after the Quran, directing Muslims in various aspects of life; the *Sahih al-Bukhari* collection is among the most renowned. The complex nature of their meanings often encompassing multiple categories of messages poses a significant challenge for manual text classification, particularly as data volume grows. In this study, the content of the hadith often includes multiple message types, such as recommendations, prohibitions, and general information. This research aims to evaluate an automated classification system for Indonesian translations of *Sahih al-Bukhari* hadith, categorizing them into three classes: Information, Recommendation, and Prohibition. The study is motivated by the vast number of hadith, which requires significant time and deep understanding for people to grasp the core message of each one. This classification system is intended to facilitate the identification of primary messages, thereby making the processes of searching, studying, and understanding hadith more effective and efficient. IndoBERT is employed to generate contextual vector representations capable of capturing deeper semantic meaning, while Logistic Regression is selected for its efficiency and stability with high-dimensional data. Evaluation is conducted using a train-validation-test split approach, alongside accuracy and macro F1-score metrics. The study achieved an average F1-score of 67.43%, demonstrating that the combination of IndoBERT and Logistic Regression yields strong, consistent classification performance for this multi-label task.

Improved Convulational Neural Network dengan Transfer Learning dan Hyperparameter Tuning untuk peningkatan akurasi klasifikasi Citra Kanker Kulit

Ega Wahyu Andani — 2026-06-30

Skin cancer is one of the diseases that requires early detection to increase the likelihood of successful treatment. The use of artificial intelligence, particularly Deep Learning, has become an effective alternative in assisting the automatic classification of skin cancer images. However, the high class imbalance and visual similarity between lesion types in skin cancer datasets remain challenges in achieving optimal classification performance. This study aims to improve the accuracy of skin cancer image classification using an Improved Convolutional Neural Network based on Transfer Learning and Hyperparameter Tuning. The dataset used is HAM10000, consisting of 10,015 dermoscopy images across seven diagnostic classes. The architecture employed is MobileNetV2 as a feature extractor combined with a custom classification head. The training process was carried out using a two-phase transfer learning strategy, namely the backbone freezing phase and the fine-tuning phase. To address class imbalance, class weighting and data augmentation were applied, while model optimization was performed using grid search over the parameters of learning rate, dense layer size, and dropout rate. Model performance was evaluated using accuracy, precision, recall, F1-score, and Area Under Curve (AUC) metrics. The results show that the proposed model achieved a test accuracy of 85.50%, a validation accuracy of 84.75%, a macro F1-score of 83.14%, and a mean AUC of 0.94. These results indicate that the combination of two-phase Transfer Learning and Hyperparameter Tuning is capable of improving the performance of MobileNetV2 in skin cancer image classification. The contribution of this research is the development of a classification model that achieves high accuracy, is computationally efficient, and is capable of handling class imbalance in the HAM10000 dataset.

Optimasi Pemilihan Jenis Kayu Berbasis Multi-Kriteria untuk Mendukung Keputusan pada Industri Furnitur Menggunakan Metode Simple Additive Weighting

Iwan Giri Waluyo — 2026-06-30

Selecting the right type of wood is a crucial factor in the furniture industry, as it directly impacts product quality, cost efficiency, and market value. Relying on subjective selection processes can lead to inconsistent decision-making. This study aims to develop a decision support system to optimize wood selection for Sumber Rejeki Mebel using the Simple Additive Weighting (SAW) method. The wood types evaluated include Teak, Merbau, Kamper Samarinda, Meranti, and Borneo, based on four criteria: color, texture, price, and availability. Criterion weights were determined through subjective assessment by an expert—the furniture business owner, who possesses extensive experience in raw material selection—and were subsequently applied during the normalization and ranking stages of the SAW method. Data analysis results indicate that Kamper Samarinda achieved the highest preference score (0.668), followed by Teak (0.653), Merbau (0.615), Meranti (0.610), and Borneo (0.605). Implemented using PHP and MySQL, the system facilitates a faster, more consistent, and transparent evaluation process compared to manual methods. Theoretically, this research demonstrates that the SAW method effectively integrates various material quality criteria into a simple, easily implementable multi-criteria decision-making model. Practically, the developed system supports more objective raw material selection, thereby offering the potential to enhance product quality and operational efficiency within the furniture industry.

Deteksi Penipuan pada Transaksi Keuangan Digital Menggunakan Ensemble Learning: Studi Komparatif Random Forest, Gradient Boosting, dan XGBoost

Nurul Akbar Tanjung — 2026-06-30

Digital payment fraud in Indonesia has grown alongside the dramatic expansion of mobile money services, creating a detection problem that conventional rule-based systems are increasingly ill-equipped to handle. This paper examines whether a soft-voting ensemble of Random Forest, Gradient Boosting, and XGBoost can offer a more effective solution. The model was trained on the PaySim synthetic dataset, consisting of 6.36 million mobile money transactions in which fraudulent cases account for just 0.129 percent of all records. SMOTE was used exclusively on the training data to address the extreme class imbalance before model fitting. Five-fold cross-validated Grid Search determined the hyperparameter configuration for each constituent model. On the held-out test set, the ensemble achieved 94.7 percent precision, 91.3 percent recall, 93.0 percent F1-score, and 0.987 AUC-ROC figures that consistently exceeded those of any single algorithm. Examining feature contributions revealed that the sender balance difference and transaction amount carried the most discriminative weight, a finding that aligns with known fraud behavior in mobile payment datasets. A local streaming latency test across 5,000 consecutive transactions produced an average response time of 147.3 ms, with the 99th percentile remaining below the 200 ms operational threshold. Taken together, the results indicate that the ensemble approach is not only statistically superior but also practically deployable within the real-time constraints of digital banking environments.

Perbandingan Kinerja K-Medoids dan Improved K-Medoids Berbasis Crow Search Algorithm pada Klasterisasi Data Transaksi Penjualan Berdasarkan Silhouette Score dan Efisiensi Komputasi

Melinda Putri Azzahra — 2026-06-30

The development of digital transaction systems generates large amounts of data that need to be processed into meaningful information to support decision-making. One approach that can be used to analyze consumer purchasing patterns is clustering. The K-Medoids algorithm is known for its robustness against outliers; however, its iterative medoid search process leads to relatively high computational time. To address this limitation, an improved K-Medoids based on the Crow Search Algorithm (CSA) is employed, utilizing a metaheuristic optimization mechanism to determine optimal medoids. This study aims to compare the performance of the K-Medoids algorithm and the improved K-Medoids based on the CSA in transaction data clustering in terms of cluster quality and computational efficiency. The dataset used was obtained from a Point of Sale (POS) system of a fast-food restaurant and consisted of 18,814 transaction records. The research stages included data preprocessing, clustering using both methods, and performance evaluation based on the Silhouette Score and computation time. The results showed that both methods produced the same optimal number of clusters, namely K = 4. The K-Medoids algorithm achieved the highest Silhouette Score of 0.557791, while the improved K-Medoids based on the CSA obtained a Silhouette Score of 0.537240. In terms of efficiency, the improved K-Medoids based on the CSA required significantly shorter and more stable computation times than the conventional K-Medoids algorithm. These findings indicate a trade-off between clustering quality and computational efficiency, implying that the choice of method can be adjusted according to analytical requirements. The main contribution of this study is providing a comparative analysis of the K-Medoids algorithm and the improved K-Medoids based on the CSA on transaction data by jointly evaluating cluster quality and computational efficiency. The findings provide practical recommendations for selecting clustering methods according to analytical requirements and serve as a reference for future research on transaction data clustering.

Penerapan Algoritma Apriori dalam Menganalisis Pola Minat Beli Konsumen di Coffee Shop

Renaldi Nur Fahrizal — 2026-06-30

The currently intensive increase in competition within the cafe industry demands that business operators, such as Coffee shop Zecoff Tenggarong, not only focus on product quality but also gain a deep understanding of consumer behavior and buying interest patterns. This understanding is crucial for formulating targeted and sustainable business strategies. This research specifically focuses on analyzing consumer buying interest patterns at Coffee shop Zecoff Tenggarong through the identification of products that tend to be purchased together in a single transaction. To achieve this objective, the study employs a Data Mining approach using the Association Rule Mining technique. The core method implemented on the cafe's sales transaction data over a specific period is the Apriori Algorithm. This algorithm was chosen due to its effectiveness in processing large datasets and identifying frequently co-occurring itemsets. The data analysis process includes the stage of determining critical parameters: support (the frequency degree of the itemset), confidence (the strength of the causal relationship), and lift (the value of association improvement), which are collectively used to filter and generate the strongest and most relevant association rules. The empirical results of the study show that the Apriori Algorithm is highly effective in uncovering hidden purchasing patterns that are difficult to detect through conventional data analysis. The strong association rules derived from this mining process provide important and actionable information for the owner of Zecoff Tenggarong. The strategic implications of these findings include: formulating more targeted cross-selling marketing strategies (for example, recommending companion products that are certainly in demand), optimizing product arrangement (placing strongly associated items in close proximity), and increasing inventory management efficiency (ensuring that items frequently bought together are always in stock). In conclusion, this research concludes that the utilization of Data Mining technology with the Apriori Algorithm is a vital and transformative tool. It not only supports daily operational decision-making but also significantly enhances the coffee shop's competitiveness amidst a tight market rivalry.

Implementasi dan Evaluasi Performa Algoritma Naïve Bayes dalam Deteksi Dini Penyakit Diabetes

Nurhasanah Nurhasanah — 2026-06-30

Diabetes mellitus is one of the most prevalent chronic diseases worldwide and requires early detection to reduce the risk of severe complications through timely intervention. This study aims to implement and evaluate the performance of the Naïve Bayes algorithm in supporting the early detection of diabetes based on patients' health data. The study employed the Pima Indians Diabetes Dataset, consisting of 768 patient records with eight input attributes and one output attribute. During the preprocessing stage, zero values in physiological attributes were treated as missing values and replaced using the median of each respective attribute, followed by data consistency checking and dataset partitioning using the 80:20 split validation method. Model performance was evaluated using a confusion matrix with four performance metrics: accuracy, precision, recall, and F1-score. The experimental results showed that the Naïve Bayes algorithm achieved an accuracy of 88.31%, precision of 87.80%, recall of 90.00%, and an F1-score of 88.89%. These findings indicate that the proposed model performs well in classifying diabetes risk. The implementation of the model in a web-based application is expected to assist healthcare professionals and the general public as an early screening tool to support preliminary decision-making before comprehensive medical examination.

Penerapan Algoritma K-Means Clustering untuk Pengelompokan Pola Penjualan Sembilan Bahan Pokok pada Pusat Distribusi Berbasis Dataset Kaggle

Fitriah Fitriah — 2026-06-30

Sales data management of staple food products (sembako) is an important concern for distribution centers because stock storage decisions are still largely based on the estimation or experience of warehouse staff rather than on adequate historical data analysis. This condition creates the risk of overstocking slow-moving products on one hand, and stockouts of high-demand products on the other, potentially causing operational losses. This study aims to apply the K-Means algorithm to cluster staple food sales data based on sales patterns, using product type and total units sold as variables after being encoded and scaled. The dataset used was obtained from Kaggle, consisting of 1,200 sales transaction records. The research stages include problem identification, data collection, data pre-processing (cleaning, transformation, and standardization), implementation of the K-Means algorithm with k=3, result analysis, and model evaluation using the Silhouette Score. The clustering process was carried out using Python libraries in Google Colaboratory. The results show that all sales data were successfully grouped into three clusters labeled -1, 0, and 1, namely cluster -1 (not in demand), cluster 0 (less in demand), and cluster 1 (in demand). Cluster 1 dominates with 930 data points (77.5%), cluster 0 contains 269 data points (22.4%), while cluster -1 contains only 1 data point (0.1%), indicating an outlier among products with very low sales volume. These findings demonstrate that the K-Means algorithm is effective in identifying sales patterns and can be used as a basis for decision-making in inventory management strategies at distribution centers, particularly in determining storage priorities based on product demand levels.

Analisis Perbandingan Kinerja Cloud Amazon Web Services dan Google Cloud Platform untuk Learning Management System Menggunakan Metode Analytical Hierarchy Process

Reza Maulana — 2026-06-30

The digital transformation of education demands scalable and reliable LMS infrastructure, yet institutions often struggle to choose the right cloud platform because decisions rely only on catalog prices or features without empirical performance evidence. This study compares the performance of Amazon Web Services and Google Cloud Platform at the Infrastructure as a Service layer through five stages: provisioning two equivalent virtual machines, deploying Moodle with its database and monitoring stack via Docker containerization, executing JMeter load tests at 100, 250, and 500 concurrent users, collecting performance metrics, and evaluating them with the Analytical Hierarchy Process across four criteria of performance, reliability, cost, and integration. GCP excels in error containment (as low as 0.12%) and OS-level stability and is 27.7% cheaper, while AWS leads in throughput up to 11.7 requests per second with consistent maximum response times. AHP yields scores of 0.768 for AWS and 0.762 for GCP with a consistency ratio of 0.041. The contribution of this research is the first integrated evaluation framework combining empirical JMeter load testing on LMS workloads with multi-criteria AHP decision making, together with a portable Docker-based testing architecture replicable on both platforms.