Normal Multivariate Based Clustering of Regencies in East Java Province-Indonesia

Abstract

One of the adverse effects of development in Indonesia is the incline of welfare inequality. Particularly
in East Java Province, we can identify this condition among regencies and cities. In this thesis I intend to make
clusters among the regencies with regards to their welfare indicators. This may help the government in giving
development priority program among the regencies for reducing welfare inequality. I used the model-based
clustering method to overcome over-lapping problem found in the welfare data. Based on the Bayesian
Information Criterion, the most fitted cluster model is a three-cluster model with diagonal distributions, variable
volumes, equal shapes, and coordinate axes orientations. The first cluster, the low welfare cluster, consists of
twelve regencies. The second one is the middle welfare cluster consisting fifteen regencies. The third class, the
high welfare cluster, has twelve regencies too. Accordingly, I suggest that the government gives priorities on the
twelve low welfare regencies, particularly in clean water accessibility, literacy, child delivery helping by medical
doctors, poverty elevation, sanitation, and year of schooling.
Keywords: Welfare Indicators, Model Based Clustering, Expectation Maximization, Bayesian Information
Criterion.