Comparison of K-Means and K-Medoids Algorithms for Clustering Poverty Data in South Sumatra Using DBI Evaluation

Abstract

This research focuses on the implementation and comparison of the K-Means and K-Medoids algorithms that function as poverty data clustering in South Sumatra Province, the poverty data is taken from the Central Statistics Agency of Indonesia (BPS Indonesia). This research also aims to analyze the poverty level in South Sumatra Province by including additional variables such as average years of schooling and per capita expenditure in the community in each regency or city in South Sumatra Province. Data clustering is done by both algorithms and then the performance value is Evaluated using Davies Bouldin Index DBI shows that K-Means gives better results, with a lower DBI value (0.204 at K=5) while K-Medoids has a DBI value of 0.239 at K=5, which indicates more compact and separated clusters. The superiority of K-Means is due to the homogeneous and minimal outlier characteristics of the dataset, which makes the centroid approach more optimal than medoids in K-Medoids. With these results, K-Means was chosen as the best algorithm for clustering poverty data in the region. The use of the K-Means algorithm produces a pattern in clusters related to education, economic inequality, and poverty distribution in various regions in South Sumatra. This implementation provides insight into how data clustering techniques can be applied to socio-economic data to provide policy makers in a region with information about the region, especially information about poverty-stricken areas.

Keywords

• k-means
• k-medoids
• poverty data
• clustering
• davies bouldin index (dbi)