Klasterisasi Indikator Data Kemiskinan Kabupaten/Kota di Indonesia Menggunakan K-Means dengan Normalisasi Z-Score dan Evaluasi DBI

Elfonda Dandi Kurnia Putra (Universitas Semarang, Semarang)
Galet Guntoro Setiaji (Universitas Semarang, Semarang)
Ahmad Rifa’i (Universitas Semarang, Semarang)
Astrid Novita Putri (Universitas Semarang, Semarang)

TIN: TERAPAN INFORMATIKA NUSANTARA
Vol 7 No 3 (2026): August 2026

This study aims to optimize the clustering of poverty-level data in Indonesia using the K-Means clustering method with a Z-Score normalization approach and evaluation using the Davies-Bouldin Index (DBI). The main problem in clustering poverty data is the difference in scales among variables, which can affect the clustering results. Therefore, data normalization using the Z-Score method was performed to equalize the scales among variables. The data used in this study are secondary data consisting of 514 regencies/cities in Indonesia entitled “Classification of Poverty Levels in Indonesia,” obtained from the Kaggle repository. The dataset was compiled by Ermila, Marcelo, and Faisal Dino Bahtiar and was last updated three years ago, specifically in 2023, as stated on the website. The data consist of poverty indicators that have undergone a variable selection process, in which variables that are irrelevant or have similar meanings and may potentially cause bias were excluded from the analysis. The clustering process was conducted using two scenarios, namely K=2 and K=3, to compare the clustering results. The results show that the K-Means method can effectively cluster the data based on poverty-level categories. The evaluation using the Davies-Bouldin Index shows DBI values of 0.9417 for K=2 and 2.706 for K=3. Based on the comparison of the clustering results, the K=2 scenario produced a lower DBI value and was therefore used as a reference for determining the classification levels. The contribution of this study lies in the development of a clustering process that integrates indicator selection, Z-Score normalization, the K-Means algorithm, and Davies-Bouldin Index evaluation across 514 regencies and municipalities in Indonesia to obtain regional groupings based on similarities in the characteristics of poverty indicators. Thus, the approach developed in this study can produce more optimal clustering results and can be used as a basis for supporting decision-making in poverty alleviation efforts effectively and appropriately.