School Accreditation Prediction Based on Literacy and Numeracy: Ordinal Logistic Regression vs KNN

Abstract

School accreditation in Indonesia has traditionally relied on administrative inputs and institutional documentation, which often fail to capture the actual quality of student learning. In contrast, the National Assessment provides direct evidence of student literacy and numeracy outcomes, offering a more objective and outcome-based measure of educational quality. Leveraging these results as predictors for accreditation rankings is therefore crucial, as they reflect the competencies most relevant to effective learning delivery. This study aims to develop and evaluate classification models for school accreditation rankings using literacy and numeracy results as predictor variables. The dataset consists of secondary data from the 2023 and 2024 National School Assessments, covering 789 schools across four provinces: DKI Jakarta, Yogyakarta, Bali, and Banten. Two methods were applied, Ordinal Logistic Regression and K-Nearest Neighbors (K-NN) under two scenarios: with and without class imbalance handling. To address imbalance, two techniques were employed: Synthetic Minority Oversampling Technique (SMOTE) and Class Weight. The results indicate that K-NN consistently outperformed Ordinal Logistic Regression in both scenarios. On data without imbalance handling, K-NN achieved Accuracy, Precision, Recall, and F1-Score of 0.803, 0.705, 0.587, and 0.619, respectively. with imbalance treatment using SMOTE, the values were 0.753, 0.619, 0.686, and 0.644. While class balancing did not significantly improve overall accuracy, it enhanced the model’s ability to recognize minority classes. These findings highlight the strong relationship between literacy and numeracy outcomes and school accreditation status, demonstrating that outcome-based measures can complement traditional accreditation instruments, and that conventional statistical approaches are still relevant for modeling school accreditation.