Performance Evaluation of Clustering Algorithms for Grouping Regencies and Cities in Kalimantan Based on Socio-Demographics

Abstract

Cluster analysis is an important data exploration technique that aims to identify patterns or characteristics of data, which, in practice, are used to support evidence-based policy making, one of which is used in population and regional development. However, choosing the wrong clustering method can produce useless analysis results. Therefore, this study proposes a comparative study to determine the performance of various clustering algorithms, comparing the performance of hierarchical clustering algorithms (single linkage, complete linkage, average linkage, centroid, ward) and non-hierarchical algorithms (k-means) in regencies and cities in Kalimantan. The study uses secondary data taken from the official websites of the Central Statistics Agency of the five provinces in Kalimantan, where the data to be clustered consists of 56 regencies and cities with eight socio-demographic variables. The analysis results show that all hierarchical methods produce robust and consistent clusters compared to non-hierarchical methods. This decision is based on the silhouette score value 0.701, which is 0.308 better than k-means. This shows that hierarchical methods can clearly identify regencies and cities in Kalimantan based on their socio-demographic factors.