Abstract

Poverty in Indonesia remains a persistent challenge, and Sulawesi Tengah stands out as one of the country’s poorest provinces despite national progress. Standard poverty measurement, based on household expenditure, often masks disparities within households, while individual-based measures risk ignoring shared conditions such as assets, dependency, and housing. This study addresses these limitations by integrating individual- and household-level perspectives and adopting a multidimensional approach to poverty analysis.

Using 27,205 individual and 7,530 household observations from the National Socioeconomic Survey (SUSENAS) 2024, the research applied a three-stage framework: factor analysis, clustering, and predictive modelling. Two data representations were compared: latent factors from Factor Analysis of Mixed Data (FAMD) and the full set of original socioeconomic variables. Factor analysis revealed that poverty risks at the individual level are shaped by human capital, behaviours, and digital access. In contrast, at the household level, they are driven by dependency burden, education, and labour type. Clustering highlighted distinct profiles, with large dependent households at the highest risk and educated professional households the most secure, with these segments communicated more clearly when using latent factors. Predictive modelling showed that household-level Random Forests using all variables offered the most robust balance of accuracy (AUC = 0.875) and interpretability, while individual-level results were inflated by repeated household attributes. Variable importance confirmed the central role of dependency, education, housing quality, asset ownership, and financial/digital inclusion.

The study demonstrates the value of combining dimensionality reduction, segmentation, and prediction to capture the complexity of poverty. Findings emphasise the need for multidimensional targeting policies that address dependency, education, material deprivation, and digital and financial inclusion, with household-level Random Forests providing a practical tool for localised poverty reduction.