Algorithm Comparison of Hierarchical and Non-Hierarchical Clustering Method in Grouping Regional Poverty Variables

Abstract

One of the objectives of the main Sustainable Development Goals (SDGs) is to end poverty in all forms. Although West Sumatera Province occupies ranking seventh lowest national in poverty, there is an increase amounting to 0.11 percent in September 2022 compared to March 2022. This shows the complexity of the poverty problem in the region. The Provincial Government needs to understand the poverty situation by grouping it based on characteristics in each region. This is a strategic step so that poverty reduction policies can be developed on target and efficiently according to the conditions of each region. This study aims to investigate Clustering methods, namely a non-hierarchical method represented by K-means, Fuzzy C-means, and K-medoids also the hierarchical method, represented by Divisive Analysis (DIANA) and Agglomerative Nesting (AGNES) with complete linkage, average linkage, single linkage, and Ward’s method, to group regencies/cities and compare the performance of the Clustering methods used, to get the best method using Davies Bouldin Index and Dunn index. The results of this research indicate that the divisive analysis method and agglomerative nesting, especially in complete linkage, single linkage, and Ward’s method is the best Clustering method. This method works optimally when the number of clusters is equal to 3. It is hoped that our findings can support policies that are right on target and efficient in efforts to overcome poverty in West Sumatera.