Utilizing Machine Learning and Explainable AI for Assessing Income Poverty Risk: Evidence from EU-SILC 2023 in Slovakia

Received: 15 December 2025 | Revised: 14 January 2026 | Accepted: 19 January 2026 | Online: 4 April 2026

Corresponding author: Silvia Komara

Abstract

Machine Learning (ML) methods, driven by advances in computational power, have become indispensable tools in contemporary economic research. Unlike traditional statistical models that primarily emphasize inference and hypothesis testing, ML techniques prioritize forecasting performance and the identification of complex nonlinear relationships among variables. However, many high-performing ML algorithms, particularly ensemble and deep learning models, operate as "black boxes", rendering the contribution of individual predictors difficult to interpret. This lack of transparency raises concerns about interpretability, which are key aspects in policy-oriented analyses. Therefore, the integration of Explainable Artificial Intelligence (XAI) techniques has become crucial for bridging the gap between predictive accuracy and meaningful understanding. In this study, we assess and predict income poverty risk in Slovakia using microdata from the European Union Statistics on Income and Living Conditions (EU-SILC) dataset for 2023. To achieve that, we compare the performance of Logistic Regression (LR), Random Forest (RF), and Extreme Gradient Boosting (XGBoost), while for the model transparency, the Shapley Additive Explanations (SHAP) values are employed. This framework enables the development of models that maintain strong predictive performance while providing clear, policy-relevant insights into the underlying drivers of At-Risk-of-Poverty (AROP).