Implementation of Machine Learning and Its Interpretation for Mapping Social Welfare Policy in Indonesia

Abstract

This research leverages data from the 2022 Early Socio-Economic Registration (Regsosek) activity to develop a machine learning model capable of predicting family expenditure levels based on the Proxy Mean Test (PMT) with high accuracy. By integrating the SHAP (SHapley Additive exPlanations) method for model interpretation, we identify the contributions of socio-economic features to expenditure predictions and link them to relevant social assistance programs. We compare two regions, Kulonprogo Regency and Yogyakarta City, representing varying poverty levels, and identify unique characteristics influencing family welfare in each area. The results highlight that effective policy interventions must be tailored to the unique characteristics of each region and family, taking into account dimensions such as housing, education, income, and community expenditures. This research provides valuable insights for policymakers, demonstrating that successful poverty alleviation policies are data-driven and adaptable to the diverse socio-economic realities across regions.