Bulletin of Computer Science Research

Sistem Informasi Penerimaan Siswa Baru Terintegrasi Berbasis Web Menggunakan Extreme Programming

2026-08-27T15:06:51+00:00

The new student admission process in many vocational high schools is still carried out manually, causing accumulated paper files, recording errors, slow data retrieval, and delayed reporting. Prior studies generally stop at digitising the registration form and have not integrated the entire admission service chain into a single system. This study aims to design, build, and test a web-based admission information system that integrates seven service stages sequentially, namely public information delivery, online account registration and enrolment, payment proof upload, payment verification, online selection test, self-service admission announcement based on the registration number, and recapitulation as well as cross-period comparison reporting. The novelty lies in integrating that service chain into one application with three-role access control and built-in framework security mechanisms. The system was developed using the Extreme Programming method through the planning, design, coding, and testing stages with the Laravel framework and a MySQL database consisting of eleven relational tables. Functional testing applied black box testing based on equivalence partitioning and boundary value analysis covering seventeen positive scenarios and fifteen failure scenarios, complemented by boundary input validation and verification of security mechanisms against SQL injection and XSS. The test results show that the system meets the functional needs of the admission committee and prospective students and rejects invalid input as designed, making it feasible for improving the efficiency, accuracy, and traceability of admission administration.

Implementasi Sistem Digital dalam Pengelolaan Surat Izin Keluar Karyawan dari Area Perusahaan Terintegrasi Notifikasi E-Mail

2026-09-01T08:54:00+00:00

Paper-based employee exit permits hinder approval tracing, data recapitulation, and coordination with security personnel. This study implements SIKAK Online, a web-based work-area exit permit system integrated with email notifications at PT Vision Ease Asia. Development followed a prototyping approach comprising requirements communication, rapid design, prototype construction, user evaluation, and refinement. Requirements were gathered through observation, interviews, and document review, and the application was developed using PHP and MySQL. Evaluation comprised black-box functional testing, checks covering ten security aspects, and a 20-item Likert-scale questionnaire administered to 46 respondents. The test summary reported that all 24 functional scenarios passed and that security checks produced responses consistent with the mechanisms examined. Of 920 possible item responses, 916 valid responses yielded an actual score of 4,065 out of an ideal score of 4,580, equivalent to 88.76%; perceived usefulness obtained the highest indicator score, 90.43%. The questionnaire results describe users’ assessments of the system. This study contributes an implemented exit–return permit workflow integrating submission, supervisor approval, email notifications, monitoring by security personnel, reporting, and activity logs. Its empirical contribution is an initial evaluation of functionality, security mechanisms, and user perceptions to inform the development of traceable permit administration in a manufacturing setting. The findings support an initial evaluation of digital permit administration, with further research needed on service-time measurement, instrument validation, and more extensive security testing.

Segmentasi Permintaan Pelanggan Paket Dekorasi Wedding Menggunakan K-Means Berbasis Web

2026-09-04T14:40:37+00:00

This study aims to analyze customer demand segmentation for wedding decoration packages using a web-based K-Means algorithm at PT. WPD. The research problem is the manual processing of historical customer data, which requires considerable time and may cause errors in grouping. The dataset consists of 250 historical customer records randomly selected from 500 transaction records for the January–December 2025 period to obtain a representative sample and accelerate model testing without reducing the variation in customer characteristics. The modeling attributes include package price or budget, decoration theme, number of add-ons, and venue scale, with categorical attributes (decoration theme and venue scale) converted into numerical values through label encoding before all numerical attributes were normalized using Min-Max Normalization. The research stages consist of observation, interviews, literature study, data collection, preprocessing, normalization, cluster determination (K=3) using the Elbow method based on the Within-Cluster Sum of Squares (WCSS) value, K-Means implementation, web-based system implementation, and testing. The analysis produced three customer segments: an economy/budget segment containing 60 customers, a middle/personalization segment containing 50 customers, and a premium/exclusive segment containing 140 customers. The premium segment is the largest group and is dominated by customers choosing the Classic International Elegance package with high budgets, extensive add-ons, and large-scale venues. The web-based system successfully automates data processing and clustering and provides analytical results to support evaluation and marketing strategies. A questionnaire involving 20 respondents on user acceptance and perceived usefulness of the system resulted in a score of 88%, categorized as very strong, reflecting user acceptance of the system rather than the accuracy of the K-Means algorithm.

Klasifikasi Tingkat Kepuasan Tamu Terhadap Pelayanan Hotel Menggunakan Metode Decision Tree

2026-09-01T08:45:52+00:00

In the highly competitive era of the modern hospitality industry, service quality, cleanliness, and the completeness of facilities are the main pillars in maintaining guest loyalty and business reputation. Hotel NICI, as one of the accommodation service providers, faces challenges in extracting valuable insights from the accumulation of large-volume guest feedback survey data. This study aims to systematically classify the level of guest satisfaction with Hotel NICI's services using a Decision Tree machine learning algorithm based on survey data. The dataset used covers various evaluation dimensions, particularly guests' perceptions of staff service quality, the cleanliness level of hotel areas, and the functionality of the facilities provided. To produce a valid model, the raw data first undergoes a series of rigorous preprocessing stages. These stages include handling missing values using imputation methods, normalizing nomenclature or column names for data consistency, as well as[1] converting qualitative textual responses into structured categories, namely 'satisfied', 'fairly satisfied', and 'dissatisfied'. To determine the global sentiment label, a majority voting approach is applied to consolidate the three satisfaction dimensions into a single target variable representing the overall guest satisfaction category. Subsequently, these categorical features along with the target variable are numerically encoded using the label encoding technique, then divided into training data (training set) and testing data (testing set) with a proportional ratio of 80:20. The Decision Tree Classifier model is trained by optimizing the tree depth parameter to prevent overfitting.

Analisis Korelasi Dimensi End User Computing Satisfaction (EUCS) terhadap Kepuasan Pengguna Internal Website Pemerintahan Daerah

2026-09-16T08:36:30+00:00

Local government websites are usually evaluated from the perspective of the public as external users, whereas the perceptions of internal users who are close to the information provision process remain rarely examined, particularly for provincial communication and informatics office websites; prior studies on government websites also tend to involve the public as respondents or stop at describing satisfaction per dimension without testing which dimension is most strongly associated with overall satisfaction. This study aims to measure the satisfaction of internal users with the South Sumatra Provincial Communication and Informatics Office (Diskominfo) website and to identify the End User Computing Satisfaction (EUCS) dimensions most strongly associated with overall user satisfaction. Data were collected through a five-point Likert questionnaire from 80 purposively selected Diskominfo employees and analyzed using validity testing, reliability testing, mean analysis, and Pearson correlation. All items were valid and all dimensions were reliable (Cronbach’s Alpha 0.779–0.842). Mean scores of the five dimensions ranged from 3.00 to 3.15 (moderately satisfied), with Ease of Use the highest (3.15) and Accuracy the lowest (3.00). Correlations with overall satisfaction were strongest for Format (r = 0.78) and Content (r = 0.71), while Ease of Use (r = 0.45), Accuracy (r = 0.40), and Timeliness (r = 0.32) showed lower correlations. These findings indicate association rather than causation and apply only to the study sample. The contribution of this study is to provide empirical evidence on the EUCS dimensions most strongly associated with internal user satisfaction on a provincial communication and informatics office website, providing a basis for prioritizing improvements in display quality and content completeness, and for strengthening the verification of information accuracy.

Sistem Pembayaran Siswa Berbasis Web dengan Metode Web Engineering untuk Digitalisasi Administrasi Sekolah

2026-09-17T07:13:47+00:00

The administration of student payments at SMK Muhammadiyah 03 Palembang still relies on administrative ledgers and Microsoft Excel, so payment data is not integrated into a single database, data retrieval and report recapitulation take a relatively long time, paper-based payment receipts are prone to damage or loss, and students and parents cannot independently monitor their billing and arrears status. This study aims to develop a web-based student payment system that integrates student data management, invoice issuance, payment transactions, and reporting within a single centralized database using role-based access control. This study applies a descriptive qualitative approach with data collected through observation, interviews with the school payment officer, and a literature review. System development applies the Web Engineering method at the communication, planning, modeling, and construction stages, while the system design is modeled using use case, class, and activity diagrams. The system is implemented using the PHP programming language with the Laravel framework, an SQLite database, Tailwind CSS, and JavaScript. The resulting system provides five access rights, namely secretary, treasurer, student, parent, and principal, and is able to manage tuition contribution fees, registration fees, and re-registration fees through two channels, namely cash recording by the treasurer and bank transfer with proof upload and treasurer validation. Black box testing of 14 functional scenarios across all user roles produced a conformity rate of 100% with a valid status. The novelty lies in integrating three payment types, two payment channels, and five user roles, including parents and the principal, in one unified system.

Konsolidasi Pelaporan Multi-Divisi pada Sistem Informasi Eksekutif Berbasis Web Menggunakan Rapid Application Development

2026-08-24T17:50:29+00:00

Executive decision making requires concise, up-to-date, and consolidated information, whereas operational reporting in energy support service companies is generally still scattered across divisions, so the information has to be collected and checked manually before it can be used. This research aims to design and build a web-based Executive Information System that consolidates the reports of four operational divisions, namely General Affair, Human Resource Development, Engineering Rig & Development, and Transport, into a single shared reporting structure and presents them as executive indicators. The system was developed using the Rapid Application Development method in four phases: requirements planning through observation, interviews, documentation, and literature study; user design producing process models, database structure, indicator mapping, and twelve interface wireframes; construction using the PHP programming language and a MySQL database iteratively through prototypes; and cutover consisting of final testing, user training, and system implementation. This research offers three contributions: (1) the design of a lightweight consolidation layer that maps the heterogeneous indicators of four divisions into a single shared reporting structure, so that the benefits of an Executive Information System can be obtained without building a data warehouse; (2) the application of Rapid Application Development in the Executive Information System domain, which has so far been dominated by sequential approaches, with the executive involved directly in every prototyping cycle; and (3) a tiered presentation pattern from summary to detail combined with role-based access control, so that inter-divisional data confidentiality is preserved when the reports are consolidated. The resulting system implements fifteen functional requirements across twelve interface pages and shortens the executive information retrieval flow from five manual cross-division steps into a single dashboard access, while reducing the number of information sources that must be opened from four divisional report files into one integrated interface, equivalent to an 80% reduction in the number of steps and a 75% reduction in the number of information sources. Black box testing of nine scenarios representing all functional requirements produced valid results in every scenario (100%), therefore the system is declared functionally valid within the tested prototype scope. The limitations of this research are that the system still runs on a local network, data updates depend on divisional user input, and advanced analytics such as trend analysis, forecasting, and anomaly detection are not yet available; these become the agenda for the further research.

Pengembangan Sistem Monitoring Data Keuangan Berbasis Web Menggunakan Metode MOORA

2026-08-13T12:13:55+00:00

Financial data monitoring is a crucial aspect of vocational high school administration. SMK Ma’arif 1 Temon still performs financial recording and recapitulation manually and semi-computerized using office applications, potentially causing data duplication, reporting delays, and weak long-term decision-making. This research aims to develop a web-based financial data monitoring system and apply the Multi-Objective Optimization on the Basis of Ratio Analysis (MOORA) method to rank student payment priorities. Software development uses the System Development Life Cycle (SDLC) waterfall model, while the decision model uses MOORA with three objective criteria: SPP payment (C1, benefit, weight 0.5), arrears (C2, cost, weight 0.3), and SPP payment timeliness (C3, benefit, weight 0.2). The result is a web-based financial monitoring system prototype. MOORA calculation shows that Alternative-5 and Alternative-9 obtained the same highest optimization value of 0.1854, while Alternative-16 obtained the lowest value of -0.0225. This negative value is mathematically valid as a relative preference score, not a monetary amount. Black-box functional testing shows that all 11 main features are successfully implemented. This system is expected to help schools monitor, recapitulate, and make financial decisions more effectively, efficiently, and measurably.

Implementasi Optical Mark Recognition dengan Pendekatan Template-Based untuk Koreksi Otomatis Lembar Jawaban Pilihan Ganda

2026-09-17T17:13:59+00:00

Manual checking of multiple-choice answer sheets is time-consuming and labor-intensive, particularly when a large number of answer sheets need to be examined. Optical Mark Recognition (OMR) can be used to automate the image-based reading and correction of answer sheets. This study implements a computer vision-based OMR system that utilizes a blank answer sheet template as a reference for identifying mark changes in answer-choice areas. The system was developed using the Python programming language and the OpenCV library through several stages, including preprocessing, grayscale conversion, adaptive thresholding, contour detection, perspective correction, image alignment with the template, answer-area segmentation, answer identification, answer correction based on the answer key, and score calculation. The system was designed to recognize six output classes: A, B, C, D, E, and BLANK. Evaluation was conducted using 50 answer-sheet images, each containing 60 questions with five answer alternatives, resulting in 3,000 answer decisions and 15,000 answer-choice areas being analyzed. All 50 answer sheets were successfully processed during the validation stage. The evaluation results showed that 2,863 predictions were correctly identified and 137 predictions were incorrect, yielding an accuracy of 95.43%, macro precision of 95.62%, macro recall of 94.53%, and macro F1-score of 94.92%. Specific evaluation of the BLANK class achieved a precision of 100%, recall of 85.71%, and F1-score of 92.31%. These results indicate that the computer vision- and template-based OMR approach can be used to automatically read and correct answer sheets with good performance. Nevertheless, 18 answers that were actually blank were identified as one of the A–E options. This condition indicates that visual variations in answer areas can still affect the identification process; therefore, optimization of the template-comparison mechanism and testing under more diverse image conditions are required to improve the robustness of the system.

Sistem Pendukung Keputusan Peringatan Dini Kualitas Udara Yogyakarta Berbasis LSTM, SHAP, dan Fuzzy Tsukamoto

2026-08-13T11:31:11+00:00

The escalating urban air pollution in Yogyakarta City, particularly fine particulate matter (PM2.5), demands an air quality monitoring system that is not merely reactive, but predictive and capable of guiding public policy mitigation. This research proposes an early warning Decision Support System (DSS) for air quality by integrating Multi-Output Long Short-Term Memory (LSTM), Explainable Artificial Intelligence (XAI) using SHapley Additive exPlanations (SHAP), and Tsukamoto Fuzzy Logic. The Multi-Output LSTM network simultaneously forecasts six Air Pollutant Standard Index (ISPU) parameters based on historical sequences from the Air Quality Monitoring System (AQMS) station during 2021–2024. The SHAP method dismantles the black-box nature of the neural network to uncover atmospheric causality dynamics, while the Tsukamoto Fuzzy inference engine transforms raw numerical predictions into hazard alert levels and structured public health recommendations. Forecasting evaluation demonstrates high model accuracy, achieving Root Mean Square Error (RMSE) metrics of 4.82 for SO2, 1.86 for CO, 6.48 for PM10, 9.50 for PM2.5, 7.33 for O3, and 11.55 for NO2. SHAP feature attribution validated via statistical cross-correlation analysis confirms that PM2.5 concentration surges are significantly driven by precursor NO2 emissions with a 1-to-2-day lag (r = 0.2701, p < 0.001) through secondary photochemical reactions. DSS testing across six extreme weather scenarios proves the adaptive capability of the system in delivering proportional alerts. However, the system exhibits limitations including reliance on a single monitoring station and stochastic meteorological uncertainty requiring careful policy calibration.

Implementasi Pendekatan Ensemble Deep Learning untuk Klasifikasi Citra Ilustrasi Asli dan Hasil Generative AI

2026-08-15T16:15:59+00:00

The rapid development of Generative Artificial Intelligence (AI) has increasingly enabled the creation of digital illustrations with high visual similarity to human-created artworks, making identification more difficult and raising challenges related to originality and copyright protection. However, research on Generative AI image detection has largely focused on photographic images or real-world objects, while the classification of digital illustrations with artistic characteristics and high visual similarity remains limited. This study aims to implement and evaluate an automated classification method to distinguish original illustrations from Generative AI-generated illustrations. The proposed approach employs Ensemble Deep Learning with a Soft Voting technique that combines MobileNetV2, ResNet50, and DenseNet121. The modeling process follows the CRISP-DM methodology and is evaluated using 5-Fold Stratified Cross Validation. Internal validation achieved an accuracy of 96.00% and an AUC of 99.06%. In external testing, accuracy reached 87.60% with an AUC of 95.09%, while individual-sample testing achieved an accuracy of 80.00%. The decrease in accuracy on external data indicates challenges in generalizing the model to data with characteristics different from those in the training and validation sets. This study contributes by applying and evaluating a Soft Voting Ensemble for classifying anime- and manga-style digital illustrations through internal validation, external testing, and individual-sample testing. The results indicate that this approach has the potential to distinguish original illustrations from Generative AI-generated illustrations, although further improvements in the model's generalization capability are needed.

Pengembangan Aplikasi Pengenalan Budaya Papua Berbasis Augmented Reality pada Platform Android

2026-05-29T23:18:12+00:00

Technological developments and globalization have contributed to a decline in younger generations’ interest in recognizing and preserving local culture, including Papuan culture, which is characterized by its diversity of tribes, languages, and traditions. Conventional cultural learning media are considered less effective in attracting users’ attention in the digital era. This study aims to develop an Augmented Reality (AR)-based application for introducing Papuan culture on Android devices as a more interactive and engaging learning medium. The research employed the Waterfall method consisting of requirements analysis, system design, implementation, and testing stages. The application was developed using Unity 3D and Vuforia SDK with a marker-based Augmented Reality approach to display cultural objects in three-dimensional (3D) form in real time. The results show that the application was successfully developed and all main features functioned properly according to the design, including marker scanning, 3D cultural object visualization, menu navigation, and supporting audio based on black box testing. The contribution of this study lies in developing a cultural introduction medium that integrates Augmented Reality technology with interactive visualization to enhance users’ learning experience and support the preservation of Papuan culture through a digital approach that is more adaptive to technological advancement.

Perbandingan Kinerja Algoritma Random Forest dan Xgboost Dalam Memprediksi Harga Bitcoin Menggunakan Indikator Teknikal

2026-08-09T15:21:29+00:00

The extreme volatility of crypto assets makes Bitcoin price movements notoriously difficult to predict using conventional linear forecasting approaches. This study aims to compare the performance of two machine learning algorithms Random Forest Regressor and XGBoost Regressor in projecting Bitcoin's closing price. A quantitative approach was employed using two years of historical time-series data, with the Simple Moving Average (SMA) and Relative Strength Index (RSI) serving as primary predictor features; both models were then comprehensively evaluated using absolute and percentage error metrics. Experimental results demonstrate that the XGBoost Regressor significantly outperformed the other model, achieving superior accuracy by minimizing error margins (MAE: 2327.81; RMSE: 2978.93; MAPE: 1.99%; R2: 0.4571), whereas the Random Forest model struggled to accurately capture volatile fluctuations (MAE: 2891.19; RMSE: 3619.53; MAPE: 2.53%; R2: 0.1984). Furthermore, the findings reveal that both algorithms largely disregarded momentum indicators, basing their predictions almost entirely on short-term average trend movements. In conclusion, XGBoost is empirically proven to be a far more robust and precise algorithm than Random Forest for modeling the anomaly-prone price dynamics of Bitcoin. The primary contribution of this research lies in providing empirical evidence regarding the effectiveness of the boosting paradigm over bagging for highly volatile crypto assets, as well as the direct implementation of the winning prediction architecture into an interactive Streamlit based dashboard to facilitate practical, real world investment decision making.

Rancang Bangun Sistem Keamanan Kotak Persembahan Gereja Berbasis Internet of Things (IoT) Menggunakan Metode Research and Development (R&D)

2026-10-02T09:26:17+00:00

The security of church offering boxes needs to be addressed to prevent unauthorized access and support more effective monitoring of the box condition. This study aims to design and develop an Internet of Things (IoT)-based security system for church offering boxes that can perform user authentication, control access to the lock, detect box opening conditions, and send security information in real time. The Research and Development (R&D) method was applied through the stages of requirements analysis, design, development, implementation, and prototype testing. The system combines a NodeMCU ESP8266, an AS608 fingerprint sensor, a solenoid lock, a magnetic door switch, an OLED display, and Telegram notification. The results showed that all three trials using registered fingerprints successfully granted access, while two trials using unregistered fingerprints were successfully rejected. The system also successfully delivered Telegram notifications in all five notification types, with response times ranging from 0.9 to 1.3 seconds. The lockdown mechanism was activated after five consecutive authentication failures for three minutes, forced opening detection operated successfully, and local security functions remained operational when the Wi-Fi connection was unavailable. The novelty of this study lies in integrating fingerprint authentication, forced-opening detection, a lockdown mechanism, Telegram-based control, and testing of local security functions under Wi-Fi loss in a single prototype for church offering boxes, using non-blocking firmware that prioritizes local processing (hardware-first). The contribution is the prototype design, this firmware architecture, and functional test data that can serve as a reference for developing similar systems.

Implementasi Isolation Forest dalam Pemantauan Suhu dan pH Air untuk Optimalisasi Pertumbuhan Ikan pada Kolam Budi Daya Berbasis Web

2026-08-13T12:00:31+00:00

This study aims to implement the Isolation Forest algorithm in a water temperature and pH monitoring system to enhance the effectiveness of water quality supervision and support the optimization of fish growth in aquaculture ponds at Cisauk, which manages 18 fish cages with a total population of approximately 40,000 fish of various species, namely tilapia, Pangasius (patin), gourami, and catfish (lele). One of the primary challenges faced in this environment is the fluctuation of water temperature and pH caused by environmental changes, particularly weather variables such as heavy rainfall that can drastically lower pH, high solar intensity that significantly raises water temperature, and seasonal transitions that disrupt the stability of water quality parameters. These changes often occur suddenly, potentially inducing physiological stress, reduced metabolism, impaired growth, and even increased fish mortality. The monitoring system previously used was not yet computerized, so data recording relied on manual measurement, which could not provide real-time information or detect critical conditions quickly and accurately. In this study, temperature and pH data were obtained through digital sensors integrated with a Web-based monitoring system serving as a medium for data visualization and management. The Isolation Forest algorithm was applied to automatically detect anomalies in water quality parameters by leveraging deviation patterns in the data caused by weather changes and other environmental factors. The system was developed using the Prototyping model so that the system design could be adjusted iteratively based on field user needs. The implementation results show that the Isolation Forest algorithm was successfully applied to the system. The dataset used had a temperature range of 26.10°C–33.20°C and a pH range of 6.69–9.62. Testing on 450 historical data points showed that the model successfully identified 427 normal data points and 23 anomalous data points, with an accuracy rate of 94.89%, where anomalous data were generally marked by higher temperature and pH than normal data, with an average anomalous temperature of 31.26°C and an average anomalous pH of 8.79 due to the influence of hot weather. The integration of this algorithm with the Web-based Dashboard also proved effective in providing early warnings to staff, enabling corrective actions to be taken in a timely manner. Overall, this system has proven capable of detecting water quality anomalies accurately and quickly, thereby having the potential to improve operational efficiency, maintain water quality stability, and support the sustainability of fish farming at Cisauk.

Sistem Pendukung Keputusan Hotel Bintang Empat Berbasis SAW dan Analisis Sentimen Daring

2026-08-13T11:54:22+00:00

Selecting a four-star hotel in Riau Province is a multi-criteria decision-making problem because travelers must simultaneously consider price, user ratings, facilities, location, service quality, and customer perceptions. This study aims to develop a decision support model based on the Simple Additive Weighting (SAW) method by integrating online review sentiment to generate structured and transparent hotel recommendations. A quantitative research approach was employed through problem identification, collection of alternative and criterion data, sentiment processing, decision matrix construction, normalization, weighting, and alternative ranking. Data analysis classified room price as a cost attribute, while user rating, facilities, location, service quality, and sentiment were treated as benefit attributes. The value of each alternative was normalized according to its attribute type, multiplied by the corresponding criterion weight, and aggregated to obtain the final preference score. An illustrative evaluation of five alternatives showed that A2 achieved the highest preference score of 0.9418, followed by A1 at 0.9293, A4 at 0.9240, A5 at 0.9194, and A3 at 0.9126. These findings indicate that the lowest-priced hotel does not necessarily become the best alternative because the final recommendation depends on the balanced performance of all criteria. The novelty of this study lies in incorporating online review sentiment scores into the SAW model, thereby extending conventional hotel evaluation that is commonly dominated by numerical data and aggregate ratings. Theoretically, this study strengthens the application of multi-criteria decision-making in the accommodation sector. Practically, the proposed model can assist travelers in comparing hotels more objectively and help hotel managers identify service attributes that influence their competitive position. Nevertheless, further empirical implementation requires actual hotel data, sentiment classification validation, weight sensitivity analysis, and comparisons with other decision-making methods to evaluate ranking robustness.

Prediksi Kepribadian OCEAN dari Video Wajah dengan ArcFace dan LSTM

2026-09-29T18:01:12+00:00

Questionnaire-based personality assessment can be affected by self-report bias, while a single facial image cannot represent behavioral change during an interaction. Video-based facial modeling is therefore a complementary way to estimate apparent personality, rather than a substitute for psychological assessment. This study evaluates continuous prediction of the Big Five dimensions (Openness, Conscientiousness, Extraversion, Agreeableness, and Neuroticism) using ArcFace transfer learning and Long Short-Term Memory (LSTM). The ChaLearn Looking at People First Impressions V2 dataset contains 10,000 clips. Ten frames were sampled uniformly from each video, faces were detected with MTCNN, and crops were resized to 112 × 112 pixels. A ResNet34-based ArcFace backbone produced a 512-dimensional embedding for each frame. Two LSTM layers summarized the temporal sequence, and a multi-output regression head predicted five scores in the range 0–1. Mean Absolute Error (MAE) was used for evaluation; the reported percentage was computed as (1?MAE)×100 and is termed regression accuracy, not classification accuracy. RMSprop achieved 90.52% on validation, while the configuration labeled baseline achieved 90.46% and RMSprop achieved 90.42% on the test split. With RMSprop, Agreeableness reached 90.85%, whereas Neuroticism reached 89.84%. These results indicate that sequential facial representations can support apparent personality estimation on this benchmark, although the differences between configurations are small. In practice, the system may support preliminary behavioral analysis in controlled settings, but facial scores alone are insufficient for high-stakes individual decisions. Replication and cross-population validation are needed before use beyond the benchmark.

Benchmarking Performa dan Efisiensi Respon Model AI ChatGPT dan DeepSeek Menggunakan Metode Eksperimen Komparatif

2026-10-04T18:28:49+00:00

The rapid development of Large Language Models (LLMs) has led to the emergence of various artificial intelligence platforms, such as ChatGPT and DeepSeek, which are relied upon for data processing efficiency. However, architectural differences between platforms raise issues regarding variations in response efficiency and information reliability when handling user queries. This study aims to conduct a benchmarking analysis to objectively measure the performance and response efficiency of ChatGPT and DeepSeek. The implemented solution employs a comparative experimental method through the analysis of 10,000 interaction records from a synthetic dataset available on Kaggle. Three primary evaluation metrics were measured: response speed (in seconds), response accuracy (in percentage), and user satisfaction (User Rating on a 1–5 scale). The benchmarking analysis results indicate that DeepSeek performs better than ChatGPT across the three evaluated indicators. DeepSeek processes responses with an average speed of 1.24 seconds, significantly faster than ChatGPT, which requires 3.44 seconds. In terms of accuracy, DeepSeek achieves an accuracy rate of 89.97%, whereas ChatGPT reaches 80.26%. This is consistent with user satisfaction, where DeepSeek earns a User Rating of 4.80 out of 5.00 compared to ChatGPT's 4.00. This study makes two contributions. First, it provides empirical benchmarks of the comparative performance of both AI models on three indicators within a single dataset, to support the selection of an optimal language processing system. Second, it provides comparison calculations with formulas and calculation steps, so that the results can be traced and re-checked. The results apply to the analyzed data.

Rancang Bangun Aplikasi Pengingat Saat Teduh Harian Berbasis Android Menggunakan Model Waterfall

2026-10-02T07:37:27+00:00

Daily devotion is an important activity for Protestant Christians to build a closer relationship with God through prayer and Bible reading, yet it is hard to maintain consistently because of busy schedules, the lack of a dedicated reminder, and unstable mobile internet connectivity in the Teluk Selimau area. This study aims to design and develop an Android-based daily devotional reminder application named TeduhKu using the Waterfall model, which consists of five phases: requirements analysis, design, implementation, testing, and maintenance. System requirements were collected through interviews with the pastor and church administrators, questionnaires distributed to congregants, and a literature study. The application was built with Kotlin and XML layouts in Android Studio, using the MVVM architecture, an offline-first approach based on SQLite through the Room Persistence Library, AlarmManager for local alarms with Snooze and Stop functions, and Firebase Cloud Messaging as a supplementary notification channel. The novelty of this study lies in integrating daily devotional content filterable by time of day (Morning, Afternoon, Evening, Night), local alarms, a reflection journal, and supplementary notifications in a single application whose core functions do not depend on the internet. The contribution of this study is the design and implementation of a devotional support application that can run under limited connectivity. Black-box testing of 23 test scenarios shows that all scenarios were valid, giving a success rate of 100% (23 out of 23 scenarios). The testing measures only functional conformity; the impact on the regularity of congregants’ devotions has not been measured.

Implementasi RPA Berbasis Unattended Bot dan ETL untuk Otomatisasi Report Tracking Truck Industri Logistik

2026-10-08T18:03:14+00:00

The Truck Tracking Report process at PT Linfox Logistics Indonesia combines vehicle, departure-time, route, task-status, and distance-related data from a third-party tracking platform. Automating extraction alone does not resolve data-quality problems because retrieved records may still contain duplicates, empty cells, and inconsistent formats. This study implements an unattended Robotic Process Automation (RPA) workflow integrated with Extract, Transform, Load (ETL): UiPath orchestrates extraction, Python transforms the data, Google Spreadsheets serves as an intermediary store, and Power BI provides visualization. The system is evaluated through functional tests, a preparation-time comparison, and a PIECES questionnaire completed by 20 respondents. The results show that the workflow handles duplicate and empty data, reduces report-preparation time from approximately 60 to 20 minutes, passes ten designed test scenarios, and receives high user ratings. These findings apply to the dataset, test scenarios, and respondents in the organization studied; dashboard updates are scheduled and are not claimed to be real-time.

Prediksi Harga Bawang Merah Menggunakan Gated Recurrent Unit (GRU)

2026-10-06T17:09:48+00:00

Shallots are one of the horticultural commodities with high economic value and play an important role in food supply in Indonesia. Fluctuations in shallot prices in East Java Province highlight the need for forecasting methods to estimate prices in future periods. This study applies the Gated Recurrent Unit (GRU) method to predict shallot prices using a univariate time series approach. The data consist of daily shallot prices obtained from the Food Staple Availability and Price Development Information System (SISKAPERBAPO) of East Java Province, covering the period from January 1, 2015, to March 31, 2026. After data cleaning, a total of 4,108 observations were obtained. The research stages include data exploration, data cleaning, normalization using Min-Max Scaling, sequence formation, model training, and performance evaluation using Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), and Mean Absolute Percentage Error (MAPE). The experiments used an 80:10:10 ratio for training, validation, and testing data, respectively, with variations in sequence length, the number of hidden units, and the number of epochs. The best configuration was selected based on the lowest MAPE value on the test data, while also considering MAE and RMSE. The experimental results identified the best configuration as a sequence length of 30 days, 64 hidden units, 50 epochs, and a batch size of 32, yielding an MAE of IDR 270.26, an RMSE of IDR 423.46, and a MAPE of 0.7073%. The model with this configuration was subsequently used to perform three-day recursive forecasting. The forecasting evaluation yielded an MAE of IDR 518.39, an RMSE of IDR 538.34, and a MAPE of 1.43%. These results indicate that the GRU model can produce shallot price predictions with relatively low error levels under the evaluated conditions. However, forecasting performance remains dependent on historical data patterns and the model configuration used; therefore, the predictions should be interpreted with consideration of these limitations.

Pengembangan Immutable Object Storage dengan Concurrent Garbage Collection dan Read-Proxy untuk Mitigasi Ransomware

2026-10-07T10:46:58+00:00

Modern ransomware targets both operational data and backup repositories, while mutable storage can still permit physical deletion through a compromised application. This study develops and evaluates a single-node Immutable Object Storage system based on Content-Addressable Storage (CAS) and FastCDC. Its contribution is an integrated design combining chunk-level deduplication, verified reads without directory access for the public application programming interface (API), and concurrent reclamation that revalidates references and activity before physical deletion. The system separates the public API from Vault Core, derives BLAKE3 chunk identities, and manages manifests and references in BadgerDB. Concurrent Garbage Collection (GC) reclaims unreferenced chunks, while a Read-Proxy over a Unix Domain Socket (UDS) provides the API reading mechanism. A shared Store instance, read leases, an active-upload registry, a grace period, and final validation coordinate reads, uploads, manifest changes, and physical deletion. Unit and race-detector tests, a production-component configuration test, and a UDS performance harness show that principal flows operate as designed, with no errors found in Read-Proxy samples. Four uploads totaling 32 MiB achieved a 96.1961% chunk-reuse ratio and 93.6187% byte savings. With 16 concurrent clients, median Read-Proxy latency was 1108.762 ms. GC reclaimed 2,056,192 bytes, while read throughput decreased by 9.1692% when GC was active. The architecture narrows public API authority, preserves integrity verification, and provides a detection window before physical reclamation. The evaluation is limited to a synthetic dataset, one device, one FastCDC configuration, and at most 16 clients.

Pengembangan Aplikasi Pembelajaran Logika dan Penalaran Berbasis Mobile dengan Gamifikasi untuk Siswa Sekolah Dasar

2026-10-06T18:56:36+00:00

Logic and reasoning learning in elementary school can benefit from interactive media that allow students to practice sequences of instructions and receive immediate feedback. This study aims to develop LogicLand as an Android-based learning medium designed to help elementary school students practice logical thinking, reasoning, and problem-solving through block-command puzzles and gamification. Development follows the Waterfall model through literature review and requirements analysis, design, implementation, and testing. The student application uses Flutter, while the server and administrator website use Laravel, REST API, and MySQL. Its features include registration and login, a dashboard, level selection, block-command puzzles, coin and star rewards, a robot collection, and learning progress records. In the puzzle activity, students arrange instructions to move a robot across a grid to a target; the system checks the resulting position to determine level completion and rewards. Functional evaluation using Black Box Testing reports 26 scenarios: 17 for the mobile application and 9 for the administrator website. All 26 scenarios were reported successful against their expected outputs, yielding a 100% functional pass rate among the tested scenarios. The study contributes an interactive practice medium that combines logical instruction sequences, progressive challenges, and game rewards in a student application, supported by administrator-based content management and documented functional testing. Practically, LogicLand offers an alternative game-based environment for practicing logical sequences and allows additional levels to be managed through the administrator website. The reported tests establish functional conformance for the scenarios examined; the effects on students' skills and learning outcomes remain topics for future empirical research.

Implementasi Convolutional Neural Network untuk Klasifikasi Kualitas Sarang Burung Walet Menggunakan MobileNetV2 dan EfficientNet-B3

2026-10-04T16:34:48+00:00

The quality of swiftlet nests is a main factor affecting their selling price, given its high economic value. Quality assessment still performed visually and manually is prone to subjectivity and inconsistency. Therefore, a system that can classify swiftlet nest quality automatically and objectively is needed. This study develops a Convolutional Neural Network (CNN)-based classification system by comparing MobileNetV2 and EfficientNet-B3 through transfer learning with a two-phase training strategy. The dataset consists of 1,200 swiftlet nest images in three quality categories, namely high, medium, and low (400 images per category), all primary data except the high category, which comprises 300 primary and 100 secondary images. The data were split into 70% training, 15% validation, and 15% test sets, after which augmentation was applied only to the training data for both architectures, so the validation and test sets contain only original images with no shared parent images across subsets. The system is implemented as a website, Qualet, that lets users upload nest images and obtain classification results. MobileNetV2 achieved a test accuracy of 78.89% with a loss of 0.6767, while EfficientNet-B3 achieved 80.56% with a loss of 0.6954. Misclassifications in both models occurred mainly in the medium category, presumably because it is a transitional class with less distinct visual boundaries and a limited number of original images. Both architectures showed comparable performance, with an accuracy difference of 1.67 percentage points, equivalent to 3 of the 180 test images. This study contributes a comparison of two CNN architectures and the Qualet website.

Perbandingan K-Means dan K-Medoids untuk Analisis Perilaku Operasional Kunjungan Perpustakaan Sekolah Menengah Atas

2026-10-13T13:14:15+00:00

Library visit records provide information about patron behavior, yet aggregate counts do not explain differences in individual activity intensity and recency. This gap hampers senior high school library managers with limited staff in setting service priorities. This study compares K-Means and K-Medoids to analyze the operational visit behavior of the SMAN 48 Jakarta library. The dataset contains 10,050 transactions recorded between August 18, 2025 and October 7, 2026, represented as 1,033 anonymized visitor profiles. The study follows the CRISP-DM framework, constructs nine recency–frequency–temporal features, applies logarithmic transformation and Z-score standardization, and evaluates 14 configurations with two to eight clusters. K-Means with two clusters achieves the highest Silhouette score of 0.4114 and a Davies–Bouldin Index of 1.1651, whereas the best K-Medoids configuration uses five clusters with scores of 0.2825 and 1.2348. The selected partition contains 165 low-intensity visitors (15.97%) and 868 relatively more active visitors (84.03%). Ten initialization runs reproduce the membership exactly, while removing active days, which duplicates visit frequency, lowers the Silhouette to 0.3896 with an Adjusted Rand Index of 0.9705; the two-segment structure persists although part of its separation stems from double weighting. The novelty lies in an evaluation protocol for school library visit logs that combines a two-algorithm comparison, metric re-verification, initialization stability testing, and feature sensitivity analysis. Its practical contribution comprises a basis for peak-hour staff scheduling, a follow-up list of infrequent visitors, and a semester-based segment update cycle. The results describe recorded utilization and do not establish reading interest, satisfaction, or academic impact.

Analysis of Food Distribution Efficiency Clustering Based on Cost and Delivery Time Using the K-Means Algorithm

2026-10-18T08:20:49+00:00

Food distribution efficiency is an important factor in maintaining supply chain performance, particularly in controlling transportation costs and ensuring timely delivery. Inefficiencies in distribution may lead to increased operational expenses and delays in delivering food products to their destinations. This study aims to analyze food distribution efficiency by grouping distribution activities based on transportation costs and delivery time using the K-Means clustering algorithm. The research stages consist of data collection, data preprocessing, feature selection, data standardization, determination of the optimal number of clusters, implementation of the K-Means algorithm, and cluster evaluation. Transportation cost and delivery time are used as the main variables to identify distribution patterns and classify distribution activities into efficiency groups. The resulting clusters are intended to distinguish distribution activities with relatively efficient, moderate, and less efficient characteristics, providing a basis for identifying operational patterns that require improvement. Data visualization is also used to support the interpretation of clustering results and facilitate comparisons among groups. The experimental results, based on 330 simulated food distribution records, produced three clusters. Cluster 1 consisted of 110 records, with an average distribution cost of IDR 176,363 and an average delivery time of 2.83 hours. Cluster 2 contained 111 records, with an average distribution cost of IDR 338,472 and an average delivery time of 5.83 hours. Cluster 3 comprised 109 records, with an average distribution cost of IDR 535,296 and an average delivery time of 9.49 hours. The clustering evaluation yielded a Silhouette Score of 0.679 and a Davies–Bouldin Index of 0.444. This study is expected to provide an analytical framework for supporting data-driven decisions in food distribution management. The proposed approach can assist distribution managers in evaluating cost and delivery-time patterns and identifying opportunities to improve operational efficiency. The validity of the findings depends on the quality of the operational data and the evaluation of the resulting clusters.

Klasifikasi Tutupan Awan Citra MODIS Menggunakan MobileNetV2 sebagai Tahap Pra-Pemrosesan Pemantauan Produktivitas Padi

2026-07-10T08:53:26+00:00

Cloud cover is a major constraint on the use of optical satellite imagery for rice productivity monitoring in tropical regions, because it degrades surface reflectance and introduces errors into vegetation index calculations. Most cloud detection methods still rely on spectral thresholds, which struggle to separate clouds from surface objects with similar reflectance. This study develops a cloud cover classification model based on a Convolutional Neural Network (CNN) with the MobileNetV2 architecture, applied to MODIS MOD09GA imagery as a preprocessing stage before rice productivity analysis. The dataset consists of 365 daily image patches from 2024 covering Pandeglang, Lebak, and Serang Regencies in Banten Province. Initial labels were produced by a rule-based spectral auto-labeling method using Band 1 and Band 4 reflectance. The data were split 70:15:15 and enhanced with mild augmentation. On the test set, the model achieved an accuracy of 71%, with precision of 72%, recall of 85%, and an F1-score of 78% for the cloudy class. The high recall for the cloudy class suits the role of the model as a filter for cloud-contaminated images ahead of time-series analysis, rather than as a pixel-level cloud segmentation tool. These findings indicate that the lightweight MobileNetV2 architecture has potential as an automated preprocessing layer in remote sensing-based rice productivity monitoring.

Analysis of the Effect of SMOTE on the Performance of Random Forest in Classifying the Risk of Death in Heart Failure Patients

2026-10-30T13:38:17+00:00

Class distribution imbalance in the heart failure patient dataset has the potential to affect the ability of the classification model to identify patients experiencing death events. This study aims to analyze the effect of the Synthetic Minority Over-sampling Technique (SMOTE) application on the performance of the Random Forest algorithm in classifying death events in heart failure patients. The dataset used is the Heart Failure Clinical Records Dataset consisting of 299 patient data, with 12 predictor attributes and one target attribute, namely DEATH EVENT. Data processing was carried out using the Python programming language through the Google Colab platform. The research stages include data preparation, dataset division with a ratio of 80:20, application of SMOTE to training data, formation of a Random Forest model, and performance evaluation using accuracy, precision, recall, and F1-score metrics. In addition, Stratified 5-Fold Cross-Validation was applied as an additional evaluation to analyze model performance on several data divisions. The test results with an 80:20 data split showed that the model before and after applying SMOTE produced an accuracy value of 83.33%, a positive class precision of 80.00%, a recall of 63.16%, and an F1-score of 70.59%. Meanwhile, the results of Stratified 5-Fold Cross-Validation showed that the application of SMOTE increased the average recall value from 69.74% to 76.00% and the F1-score from 73.34% to 74.75%. However, the average accuracy value decreased from 83.93% to 83.60%, while the precision decreased from 77.93% to 73.74%. These findings indicate that the application of SMOTE has the potential to improve the ability of Random Forest in identifying patients with the label DEATH EVENT = 1, although it does not improve all evaluation metrics. This study is expected to provide an empirical contribution in evaluating the effectiveness of SMOTE in classifying imbalanced medical data, especially in improving the detection ability of minority classes.