Empirical Analysis of Machine Learning Approaches for Credit Risk Prediction in Nigerian Microfinance Institutions

Credit risk assessment in Nigerian microfinance institutions relies predominantly on manual officer judgment and limited borrower financial data, creating high default rates and credit access constraints. This study empirically evaluates machine learning approaches for credit risk prediction using transaction and borrower profile data from Nigerian microfinance institutions. A dataset of 18,400 loan records from three microfinance institutions in Lagos, Kano, and Enugu States, spanning 2018 to 2023, was obtained under research data sharing agreements. Features included borrower demographics, loan amount, repayment history, business sector, and social capital proxies. Five classification algorithms were evaluated: Logistic Regression, Random Forest, Gradient Boosting (XGBoost), Support Vector Machine, and an Artificial Neural Network. Models were trained on an 80-20 stratified train-test split. Class imbalance was addressed using SMOTE oversampling. Evaluation metrics included AUC-ROC, precision, recall, and F1-score. XGBoost achieved the highest AUC-ROC of 0.89, compared to 0.74 for Logistic Regression. Feature importance analysis identified repayment history and loan cycle number as the strongest predictive features, with informal social capital proxies (e.g., community group membership) showing unexpected predictive value. The study fills a documented gap in Nigeria-specific machine learning credit scoring research and demonstrates that ML-based credit models substantially outperform traditional approaches. Recommendations include piloting XGBoost-based scoring in Nigerian microfinance and addressing data privacy obligations under the NDPR 2023 through anonymisation protocols.

Keywords: credit risk prediction, machine learning, XGBoost, microfinance, Nigeria