Track: Data Analytics

Abstract

As people's demands grow, so does the need for bank loans. Every day, banks get many loan applications from customers and other individuals but not every applicant is accepted. Typically, banks execute a loan application after verifying and evaluating the applicant's eligibility, which is a time-consuming and challenging process. When examining loan applications and making credit approval decisions, most banks use their credit score and risk assessment systems. Despite this, some applicants fail to pay their bills each year, causing financial institutions to lose a substantial amount of money. In this study, Machine Learning (ML) algorithms are employed to extract patterns from a common loan-approved dataset and predict deserving loan applicants. Customers' previous data will be used to undertake the study, including their age, income type, loan annuity, last credit bureau report, Type of organization they work for, and length of employment. ML methods such as Logistic Regression, Random Forest, XGBoost, Adaboost, Lightgbm, Decision tree, Nave Bayes, and K-Nearest Neighbor were used to discover the maximum relevant features, i.e., the elements that have the most impact on the prediction output. These mentioned algorithms are compared and assessed against one another using standard metrics. Among these, Logistic Regression achieved the highest accuracy of 92%. It was also determined as the best model and performed significantly well better than other machine learning methods in terms of F1-Score, which is 96%.

Keywords: Financial Fraud detection, Loan Sanction, Machine Learning, Logistic Regression.