A Machine Learning-Based Model for Predicting Credit Facility Defaulters in Uganda
Loading...
Date
2024
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Predicting credit facility defaulters in Uganda poses a significant challenge, particularly for individuals lacking formal banking histories. This study aims to address this gap by leveraging machine learning techniques on a diverse set of financial data, including mobile money transactions, FinTech services, and traditional banking records. By developing a more inclusive creditworthiness assessment tool, we seek to enhance financial inclusion for underserved populations. Several machine learning algorithms to predict loan defaults were evaluated, including Logistic Regression, Support Vector Machines (SVM), and Extreme Gradient Boosting (XGBoost). The XGBoost model emerged as the most effective, achieving an accuracy of 95.23%, a recall of 73.32%, a precision of 94.11%, and an AUC of 0.8119. In contrast, the Logistic Regression model attained an accuracy of 89.53%, with a significantly lower recall of 43.24% and precision of 66.59%. The SVM model performed moderately, with an accuracy of 93.21% and a recall of 62.80%, but it still fell short compared to XGBoost. The findings highlight the potential of advanced machine learning models like XGBoost to significantly improve credit scoring systems. By providing a more accurate and inclusive tool for credit evaluation, financial institutions and policymakers can better identify potential defaulters, mitigate loan defaults, and foster economic growth among underserved communities