Published Paper Details:

Microfinance Institutions (MFIs) have proved to be the best support for low-income people in achieving their financial goals. However, there has been always a risk for loan defaults which can threaten the stability of MFIs. Previous techniques have less emphasized on this problem. They were much more into data analytics. Therefore, in this study we used machine learning and deep learning techniques to improve loan default prediction using a large FinTech dataset. Four machine learning and one deep learning technique was developed and compared. The results show that Logistic Regression performed best overall with a ROC-AUC of 0.7186 and recall of 0.65. XGBoost achieved high accuracy of 84% but performed poorly in identifying defaulters. The Neural Network showed competitive performance with a ROC-AUC of 0.7055 and recall of 0.58. The study concludes that Logistic Regression remains a strong and explainable baseline model for MFIs, while advanced models like Neural Networks have good potential for loan default classification.