An Exploration of Deploying Predictive Analytics for Loan Default Using Ensemble Learning Techniques

Abstract

The purpose of this article is to explore ensemble machine learning (ML) algorithms for selecting the right customers for granting loans based on certain information about the loan applicant. The profit of the bank is largely dependent on the extent to which the loans are paid back. Banks try to reduce their non-performing assets by the correct selection of customers for approving loans. In this work, ML approaches of Logistic Regression, Decision Tree, K-Nearest Neighbor (KNN) classifier, support vector machines (SVMs), and Naive Bayes classifiers are used. Further advanced bagging and boosting algorithms like Random Forest (RF), adaptive boosting (AdaBoost), Extreme Gradient Boosting (XGBoost), and Gradient Boosting algorithms are applied further to classify the loan applicants, whether their application will be rejected or accepted. Results show that the boosting algorithms perform better than the standard ML techniques. The performance of the classifiers has been determined by using the standard performance evaluation measures of accuracy, precision, recall, and F1 score. The research will help banks and financial organizations develop systems that can identify the right customers to grant loans and, in a way, enhance their profit. Different ML approaches have been considered and compared for generating accuracy in loan prediction. The results are interesting as they exhibit the efficiency of boosting algorithms over standard ML approaches in this kind of loan default problem.