Empirical Analysis of Machine Learning Algorithms for Credit Scoring in Nigerian Microfinance Banks

Credit scoring in Nigerian microfinance banks traditionally relies on collateral-based assessments and loan officer judgement, resulting in high non-performing loan ratios and financial exclusion for creditworthy low-income borrowers. This study empirically analyses the performance of machine learning algorithms for credit scoring in Nigerian microfinance banks, addressing a gap in localised algorithmic credit assessment research. A dataset of 15,000 historical loan records was obtained from three microfinance banks operating in Ogun (South West), Plateau (North Central), and Akwa Ibom (South South) States. Six algorithms are evaluated: logistic regression, decision tree, random forest, gradient boosting, support vector machine, and a multi-layer perceptron neural network. Evaluation metrics include area under the ROC curve (AUC), precision, recall, and F1-score, with stratified cross-validation applied to address class imbalance. Random forest achieves the highest AUC of 0.89, while gradient boosting yields the strongest F1-score of 0.84. Feature importance analysis identifies repayment history, mobile money transaction frequency, and social capital indicators as the three strongest predictors of creditworthiness in Nigerian microfinance contexts. The study evaluates model fairness across gender and geographic subgroups, finding mild disparities in recall that warrant attention. Findings fill a research gap by providing algorithm benchmarks grounded in Nigerian financial data characteristics. Keywords: credit scoring, machine learning, microfinance, Nigeria, random forest