<![CDATA[Academic data contains complex patterns that require appropriate clustering approaches to support informed educational decision-making. However, comparative studies that regularly evaluate various clustering methods for student academic performance, using diverse public data sets and consistent evaluation criteria, are limited. This study aims to identify the most effective clustering algorithm for modeling student academic performance by comparing three techniques: K-Means, GMM, and BIRCH, on two publicly available datasets: the Student Performance Metrics (SPM) Dataset with 16 features and 493 instances, and the Higher Education Students Performance Evaluation (HESPE) dataset with 32 features and 145 instances. Algorithm evaluation was performed using Sum of Squared Errors (SSE), Davies–Bouldin Index (DBI), Silhouette Score, and computational time. The results show that K-Means consistently provides superior clustering quality on both datasets, outperforming the other algorithms in four evaluation criteria, while BIRCH demonstrates superiority in two metrics and achieves the shortest computational time. These findings highlight that clustering effectiveness is strongly influenced by algorithm characteristics and data structure, with K-Means being more suitable for accuracy-oriented clustering and BIRCH for time-critical applications. Overall, this study contributes to educational data mining by providing comparative evidence on algorithm performance and demonstrating how methodological choices influence the interpretation of student performance patterns. In practice, institutions can choose clustering methods that best suit their needs, such as K-Means for precise academic profiling or BIRCH for rapid, large scale analysis, to help students graduate successfully.]]>
Copyrights © 2025
