Comparative Analysis of Support Vector Machine (SVM) and K-Nearest Neighbors (KNN) for Student Performance Prediction

The application of Educational Data Mining (EDM) plays a crucial role in predicting students’ academic performance to support early warning systems in higher education institutions. This study aims to conduct a comparative evaluative analysis of the predictive outputs from the Support Vector Machine (SVM) and K-Nearest Neighbors (KNN) models. The evaluation focuses on the ability of both algorithms to process factors determining student performance, represented through static features (initial inherent profiles) and temporal features (dynamics of learning behavior). Using an evaluative case study design based on the analysis of computational output documents, this study tested 96 academic data samples categorized into target classes H, L, and M. The test results demonstrated that the SVM model performed better overall, with an accuracy of 61.46%, compared to KNN, which achieved 58.33%. SVM demonstrated more stable predictive performance overall, including for the hard-to-predict H and M classes. On the other hand, KNN experienced a drastic decline in metrics due to confusion in distinguishing pattern boundaries between the H and M classes, although specifically, this algorithm performed with higher precision in detecting the L class (F1-score 0.78). In conclusion, SVM is recommended as the superior model for overall accuracy, while KNN is only relevant for specific detection tasks focused on a single category (Class L). These findings also underscore the need for future feature engineering optimization to address the high prediction bias in Classes H and M.