Evaluation of the Performance of K-means and Bisecting K-means in Clustering Indonesian Regions using Poverty Data

Clustering regions in Indonesia based on poverty data is essential for developing targeted policies. A key challenge is determining the optimal number of clusters to accurately reflect regional disparities. This study compares the Bisecting K-means and K-means algorithms, evaluating them with Sum Squared Error (SSE), David-Bouldin Index (DBI), and Silhouette Coefficient (SC). Evaluation results identified K-means with six clusters (k= 6) as the most optimal model. It achieved a low SSE (2129.37), a relatively low DBI (0.82), and a moderate SC (0.3625). This model successfully maps regions from the most prosperous to the poorest, providing a clear basis for poverty alleviation strategies. For future work, a deeper analysis of cluster members using heatmaps or boxplots is recommended. Visualizing the results on a map would also help stakeholders easily understand the spatial distribution of poverty levels. Furthermore, comparing clustering results year-on-year would be valuable for tracking regional progress.