PERBANDINGAN METODE K-MEANS DAN DBSCAN PADA ANALAISIS KLASTER MULTIVARIAT PROFIL SISWA

Abstract

In educational practice, students are often treated as a homogeneous group within a heterogeneous environment. The "one-size-fits-all" approach has proven ineffective as it overlooks the segmented needs of diverse individuals. This study aims to perform student profiling through cluster analysis to identify groups requiring specialized support or enrichment. The methodology compares two clustering algorithms: K-Means (partition-based) and DBSCAN (density-based). Prior to modeling, optimal parameters were determined using the Elbow method for K-Means and the K-distance graph for DBSCAN. Model evaluation was based on the number of resulting clusters and the Silhouette Score. The results indicate that DBSCAN outperformed K-Means with a Silhouette Score of 0.4 compared to 0.3. Furthermore, DBSCAN produced a more optimal and cohesive structure with 5 clusters, making it more interpretable for student profiling than the 8-cluster solution from K-Means. The analysis successfully identified five student profiles: High Achievers, Above Average, Moderate, Below Average, and At-Risk/Underprivileged. This study concludes that multivariate cluster analysis is an effective instrument for targeted educational intervention, particularly in prioritizing economic aid and academic mentoring for vulnerable groups without relying on demographic identity as a determinant of individual ability.