Abstract Book of the 8th World Conference on Social Sciences
Year: 2025
[PDF]
Targeting Social Assistance Beneficiaries Using Machine Learning: A Poverty Probability-Based Approach
Dr. Sahraoui Chaymae
ABSTRACT:
In many developing countries, improving the targeting of social assistance remains a persistent challenge, particularly in identifying vulnerable individuals excluded from existing programs. This study explores the potential of Machine Learning to enhance the accuracy and equity of beneficiary identification, using a poverty probability-based approach. Drawing on real-world survey data from a publicly available database, we develop a classification model trained to estimate the likelihood of household poverty based on multidimensional socio-economic indicators. The resulting predictive model is then applied to test data to simulate real-world targeting scenarios. Preliminary results suggest that supervised learning techniques can effectively identify non-beneficiary households that exhibit characteristics of vulnerability, thus reducing errors of exclusion. The study also highlights the practical implications of such data-driven methods for policy design, especially in contexts where traditional targeting mechanisms show limitations. Beyond technical performance, we emphasize the ethical considerations surrounding transparency, fairness, and acceptability of algorithmic decision-making in public welfare systems
Keywords: Algorithmic targeting, Data-driven social policy, Poverty prediction, Social protection systems, Vulnerability detection