EXTREME GRADIENT BOOSTING ALGORITHM FOR CREDIT RISK PREDICTION USING MACHINE LEARNING (A CASE STUDY OF MODIBBO ADAMA UNIVERSITY MICROFINANCE BANK)

ABSTRACT This study develops an Extreme Gradient Boosting (XGBoost) model for credit risk prediction in Modibbo Adama University Microfinance Bank. XGBoost, a high-performance ensemble machine learning algorithm, was applied to classify loan defaults by leveraging its capacity for handling non-linear relationships, missing data, and imbalanced classes. The primary objectives were to identify key factors influencing default risk, apply a predictive model, and evaluate its effectiveness for proactive credit management. Secondary data were obtained from the bank’s historical loan records, including borrower demographics, loan characteristics, repayment behavior, and account balances. The data were preprocessed, anonymized, and analyzed using Python-based tools. The methodology involved descriptive analysis, supervised model training, feature engineering, and evaluation using accuracy, precision, recall, F1-score, and ROC-AUC. Results revealed that customer account balance, loan amount, and tenure were the strongest predictors of default. The XGBoost model achieved an overall accuracy of 98%, with strong precision and recall in distinguishing performing from non-performing loans. The study recommends integrating XGBoost into the bank’s loan approval and monitoring systems, strengthening data infrastructure, and using feature importance insights to refine credit policy. Overall, the findings underscore the potential of machine learning for enhancing risk evaluation and sustaining financial stability in microfinance institutions.