Robust Clustering Analysis of Village Potential Data in South Sumatra Using MMCD-Based Mahalanobis Distance

Dwi Fitrianti (School of Data Science, Mathematics, and Informatics, IPB University, Indonesia)
Anwar Fitrianto (School of Data Science, Mathematics, and Informatics, IPB University, Indonesia)
Anang Kurnia (School of Data Science, Mathematics, and Informatics, IPB University, Indonesia)

International Journal of Advances in Data and Information Systems
Vol. 7 No. 2 (2026): August 2026 - International Journal of Advances in Data and Information Systems

Village potential data often contain outliers that can distort clustering results when classical distance measures are applied. This study evaluates the effectiveness of the Robust Mahalanobis Distance based on the Matrix Minimum Covariance Determinant (RMD-MMCD) for clustering village potential data in South Sumatra Province, Indonesia. The proposed approach is compared across K-means, K-medoids, DBSCAN, and DB-Kmeans algorithms, with performance assessed using the Silhouette coefficient, Dunn index, Davies–Bouldin index, Rand Index, and Adjusted Rand Index. The findings demonstrate that incorporating robust covariance estimation through MMCD improves clustering stability and enhances resistance to outliers while preserving the inherent matrix structure of multivariate data. Notable differences in cluster composition are observed primarily in the accessibility and transportation and public service dimensions. These results confirm the advantage of robust distance measures for producing more reliable and structurally consistent clustering solutions in village potential data.