Financial Inclusion, Energy Poverty and Food Security in Sub-Saharan Africa: Insights from Machine Learning Techniques

This study investigates the relationship between financial inclusion (FI), energy poverty (EP), and food security (FS) in Sub-Saharan Africa (SSA), with particular attention to the nonlinear role of energy conditions in shaping the effect of FI on FS. Using a panel of 30 SSA countries over the period 2010–2024, the study constructs multidimensional indices of FI and EP using Principal Component Analysis (PCA). To account for nonlinearities and complex interactions, five machine learning algorithms are considered, namely Support Vector Regression, XGBoost, Gradient Boosting, Random Forest, and Artificial Neural Networks. The results show that XGBoost provides the highest predictive performance, explaining approximately 62.49% of the variation in FS. The SHAP analysis further indicates that FI and energy conditions are among the most influential determinants of FS, although the contribution of FI is heterogeneous across observations. The XGBoost-based mediation analysis reveals a significant indirect effect of FI on FS through energy conditions, while the direct and total effects remain statistically non-significant. More importantly, the threshold analysis identifies a critical EP level of 4.044. Below this threshold, FI is associated with an improvement in FS, whereas beyond the threshold, its estimated effect becomes negative. These findings indicate that the contribution of FI to FS depends critically on prevailing energy conditions. The study therefore highlights the importance of integrating FI policies with interventions aimed at expanding access to modern energy services, particularly electricity and clean cooking technologies, in order to strengthen FS in SSA.