COMPARISON OF K-MEANS AND GAUSSIAN MIXTURE MODEL IN PROFILING AREAS BY POVERTY INDICATORS

Abstract

The Covid-19 pandemic has led to income degradation of the Indonesia population which potentially triggers poverty. According to the Indonesian Central Statistics Agency, the Province of Central Java is one of the areas that is most affected by Covid-19 especially on the economic aspect. In 2020, the percentage of poor people has increased by 0.6% from 2019. If this condition is ignored for the long term, it will have a negative impact on hampering national development. As a first step in designing a strategy for mitigating the impact of poverty, it is necessary to carry out an appropriate profiling of the areas affected on the economic aspect based on poverty indicators. This study compares the K-Means Clustering and Gaussian Mixture Model (GMM) in providing the best data grouping based on clustering indexes, including: connectivity, Dunn, and silhouette. GMM is a generalization of K-Means clustering to include information about the covariance structure of the data as well as latent Gaussian centers. We used poverty indicators data from Central Statistics Agency of Central Java, such as poverty line, percentage of poor population, poverty depth index, and poverty severity index.  The results obtained from this study indicate that the GMM gives the best results with the 3 clusters, with the number of members for the first, second, third is 10, 19, and 6 respectively.