Enhancing Credit Risk Assessment in Nigerian Banking Using Machine Learning Ensemble Models

Abstract

The Nigerian banking sector faces escalating challenges from non-performing loans, which surged from 13.2% in 2021 prediction to 16.2% in 2022, necessitating advanced credit risk assessment methods. This study develops an ensemble machine learning model integrating Logistic Regression, Support Vector Machine (SVM), and XGBoost to improve the of loan defaults. Using Kaggle’s Credit Risk Dataset, comprising 14,092 approved credits from 32,581 applications, the model was trained after preprocessing to address missing values, scale features, and encode categorical variables. The ensemble achieved 92% accuracy, 92% precision, 98% recall, and 95% F1 score, demonstrating its effectiveness in identifying potential defaulters. These results suggest that machine learning ensembles can enhance lending decisions, reduce financial losses, and improve banking stability in Nigeria. However, the use of a non-Nigeria-specific dataset limits direct applicability. Future research should prioritize localized datasets to tailor models to Nigeria’s economic context.