Using Mixed Machine Learning Approaches for Tree-Based Classification Models of Poverty Indicator Data
Abstract
This study analyzes household poverty using data collected from six remote rural villages in Thailand. The dataset includes all twelve indicators of Thailand’s Multidimensional Poverty Index (MPI), along with household characteristics such as occupation, assets and debts, and the disability and dependency status of family members. A multi-stage Machine Learning (ML) framework was applied to identify key poverty indicators and develop predictive models. First, unsupervised techniques—Self-Organizing Maps (SOM) and Frequent Pattern Growth (FP-Growth)—were used to uncover hidden patterns in the data. The analysis highlighted four indicators strongly associated with poverty status: financial burden, educational attainment, savings, and internet access. Subsequently, tree-based classification algorithms—C4.5, Random Forest, and Gradient Boosted Trees—were used to predict poverty indicators based on these features. These models were selected for their interpretability, enabling the findings to provide actionable insights for policymakers in prioritizing poverty alleviation efforts.
Authors: Phattara Khumprom, Sansiri Tarnpradab, Charoenchai Khompatraporn, May May Ho, Wanatchapong Kongkaew
Published in: International Joint Conference on Computer Science and Software Engineering (2026)