<![CDATA[This study aims to analyze the clustering patterns of equivalency education participants through a comparative clustering approach using three algorithms: K-Means (centroid-based), DBSCAN (density-based), and Louvain (graph-based). The dataset consists of 1,057 participants with numerical and categorical attributes representing heterogeneous characteristics. The research stages include data preprocessing and the implementation of clustering algorithms on the same dataset to maintain comparison consistency. Evaluation was conducted using the Silhouette Score and Davies-Bouldin Index (DBI) as internal validation metrics, as well as external validation through expert confirmation to ensure the contextual relevance of the results. The findings indicate that K-Means and DBSCAN produced a Silhouette Score of 0.040, reflecting poor cluster separation quality and the dominance of one large cluster. DBSCAN demonstrated an advantage in detecting noise; however, it was unable to significantly improve cluster separation quality in data with a high level of homogeneity. In contrast, the Louvain algorithm generated a more balanced community structure with a low imbalance ratio, making it more capable of representing relational connections among data points that are not fully captured by distance-based approaches. This study contributes through a comparative analysis across clustering approaches in the context of equivalency education, as well as through the integration of quantitative and contextual validation. The findings confirm that graph-based approaches are more adaptive for data with high homogeneity and have the potential to serve as a basis for participant segmentation to support more effective data-driven decision-making in the education sector.]]>
