LLM-Driven Identification of Digital Financial Inclusion Gaps in Developing Economies

Abstract

This study identifies the developing economies that face the largest digital-finance activation bottlenecks after basic account access has already been achieved. Using the 2024 country layer of Global Findex 2025 and merging it with World Development Indicators on internet use and mobile subscriptions, the paper constructs an activation-gap measure that captures the distance between account ownership and high-frequency digital payment use. The analytical sample contains 86 developing economies after restricting the data to national all-adult observations and to countries with complete target variables. The average country in the sample had account ownership of 58.68% but a high-frequency payment index of only 21.64%, leaving a mean activation gap of 37.04 percentage points. The empirical strategy estimated repeated five-fold cross-validated Ridge, Random Forest, XGBoost, and LightGBM models for both regression and high-gap classification. XGBoost produced the strongest regression fit, with an RMSE of 5.32 percentage points and an R² of 0.794, and it also delivered the best classifier, with an AUC of 0.863. The largest observed gaps were recorded in Ghana (64.91 percentage points), Uganda (64.61), and Zambia (60.88). SHAP-based interpretation shows that account ownership, internet use, mobile money account penetration, card-linked government payment channels, and digital skills were the most influential predictors. The results establish that the central policy challenge in many developing economies is no longer account opening alone; it is the conversion of nominal access into repeated digital payment behavior. The final diagnostic layer converts measured model outputs into country-specific policy notes, creating a reproducible bridge from predictive analytics to practical public policy design. Keywords : digital financial inclusion; Global Findex; digital payments