Multidimensional Cluster Analysis to Evaluate the Unevenness of Social Welfare Decile Change Policy in Indonesia: A Data Science Approach

Abstract

Backround: Social assistance targeting policy in Indonesia relies on a decile system within the Integrated Social Welfare Data (DTKS) to classify community welfare levels, yet this classification is frequently criticized for producing uneven exclusion and inclusion errors across regions.
Objective: This study evaluates that criticism from a data science perspective. Five province-level socioeconomic indicators from Statistics Indonesia (BPS) for 2021, namely the Human Development Index, poverty percentage, Gini ratio, open unemployment rate, and GRDP growth rate, were collected for 32 Indonesian provinces.
Method: The data were standardized and grouped using the K-Means algorithm, with the optimal number of clusters determined through the elbow method and silhouette score.
Results: The resulting multidimensional cluster classification was then compared with a single-indicator poverty-based quintile classification, used as a simplified proxy for the single-indicator logic of the government decile, not as a reconstruction of the decile itself. The results show that five clusters produced the highest silhouette score (0.2802), and 59.38% of provinces were classified differently between the two approaches, with eight provinces experiencing a severe shift of two levels or more. The Adjusted Rand Index between the two classifications was only 0.0521, indicating very low agreement.
Conclusion: These findings demonstrate substantial disagreement between multidimensional and poverty-based classification; they do not observe targeting errors, since household-level data were not accessible. The study is a province-level proof of concept; provincial aggregates cannot establish household-level exclusion or inclusion error. Multidimensional clustering may be considered as a complementary analytical tool for the further validation of welfare targeting.Keywords: data science; K-Means clustering; welfare decile; social assistance policy