Indra Maulana
Institut Prima Bangsa, Indonesia

Published : 1 Documents Claim Missing Document
Claim Missing Document
Check
Articles

Found 1 Documents
Search

A Comparative Analysis of K-Means and Gaussian Mixture Models for Student Learning Behavior Clustering Using the Davies-Bouldin Index Indra Maulana
International Journal of Education and Humanities Vol. 7 No. 3 (2026): International Journal of Education and Humanities (IJEH)
Publisher : Ilmu Inovasi Nusantara

Show Abstract | Download Original | Original Source | Check in Google Scholar

Abstract

Understanding student learning behavior remains a significant challenge in learning analytics, as accurate segmentation is essential for designing data-driven academic interventions. This study aims to compare the performance of two unsupervised clustering methods, K-Means and Gaussian Mixture Models (GMM), for grouping data on higher education students’ learning behavior and to provide empirical evidence to support the selection of more appropriate segmentation methods. The study focuses on the capacity of both methods to produce compact, well-separated, and stable clusters using academic, behavioral, and social context attributes. The dataset underwent preprocessing, including handling missing data, encoding features, and standardizing to ensure consistent analysis. Evaluation was conducted through a series of experiments using the Davies–Bouldin Index as the primary cluster quality metric, accompanied by assessments of stability, variability, and consistency of clustering results across experiments. The most important finding is that K-Means produced more compact and well-separated clusters, with a lower average Davies–Bouldin value (2.04) than GMM (2.33), and exhibited higher stability. Conversely, GMM showed greater variability despite its ability to model overlapping clusters. These differences indicate that algorithm selection affects the quality and interpretability of segmentation. The findings confirm that K-Means is better suited to learning behavior data with a relatively linear structure. This study contributes to academic discourse on clustering algorithm selection and cluster validity evaluation in educational analytics, and offers methodological guidance for future research in higher education. Implications suggest that researchers and practitioners should consider data-structure characteristics and validity metrics before using segmentation results to design targeted, data-driven academic interventions, thereby avoiding misinterpretation