Partitional Clustering of Underdeveloped Area Infrastructure with Unsupervised Learning Approach: A Case Study in the Island of Java, Indonesia

Abstract

This study attempted to identify underdeveloped areas in regencies/cities on the island of Java, Indonesia, based on a number of infrastructure indicators. An unsupervised learning approach was used to perform partition clustering with the K-Means, K-Medoids, and CLARA methods. In addition to technically obtaining clustering results and conducting a performance comparison of the three unsupervised learning methods, another objective of this research was to map the clustering results to make it easier to recognize the characteristics of the regions indicated as underdeveloped areas, which should be absolute priorities for infrastructure development. It was found that the best clustering method was the CLARA method, with a connectivity coefficient of 7.4794 and a Dunn’s index value of 0.1042. The partition clustering of regencies/cities on Java Island using the CLARA method based on infrastructure indicators resulted in 99 regencies/cities included in the cluster of areas with underdeveloped infrastructure, while 12 regencies/cities were included in the cluster of areas with developing infrastructure, and 8 regencies/cities were included in the cluster of areas with developed infrastructure.