Predicting Loan Repayment Using a Hybrid of Genetic Algorithms, Logistic Regression, and Artificial Neural Networks

Abstract

Loans are important products of financial institutions and banks. All institutions are trying to find effective business strategies to convince more customers to apply for a loan. However, some customers are unable to repay the loan after their application is approved. Therefore, many financial institutions and banks have considered some events when approving a loan. Determining whether a borrower can repay a loan is difficult. If the Financial institution, the Bank is too strict, there will be fewer approved loans, which means less profit. But if the approval is too loose, they will approve loans that default. The Machine learning classification algorithms are applied to predict loan default: Logistic Regression, Decision Tree, and Artificial Neural Networks. Accuracy, precision, recall, and ROC curve are used to evaluate the models and the results compared. We use feature selection techniques and propose models of Ensemble learning that are Logistic Regression with Decision Tree, and Logistic Regression with Decision Tree. We achieve the highest accuracy of 84.68% using the Logistic Regression with Decision Tree ensemble learning model.