A REVIEW ON COMMERCIAL BANK LOAN STATUS PREDICTION USING MACHINE LEARNING CLASSIFICATION

A credit institution needs an accurate and consistent credit risk rating system in order to operate effectively and regularly. If accurate projections are not maintained, they will not be able to conduct their organization with any acceptable levels of inaccuracy or continue in a financially viable and transparent manner. Financial organizations are finding it harder and harder to distinguish between different loan requests as default rates rise. The topic of credit risk is now a source of widespread concern on a global scale. Alternatives have been tried, but no clear solution has emerged. This study suggests a machine learning strategy that can both detect and control borrower risk. It provides a method to precisely anticipate whether or not a loan request would be approved taking into account the borrowers' financial and social history, which reduces future losses. Our supervised learning approach was employed in this experiment to examine the distinctions between former clients and clients who had defaulted on loan payments. This model will help a lender decide whether to approve or deny a loan application. Several algorithms and classifiers have been developed, and a preference has been given to one based on several metrics such AUC, F1, etc., in order to determine the combination of functions that results in the best overall measure of precision and class discrimination. The best number of components for prediction by cross-validation was determined using PCA, and the final features were found using cross-validation cross-recursive feature selection (CVEC) and subspace eversion (SV). This, by definition, enables us to use our scarce resources more effectively. Extreme Gradient Boosting Regression, Random Forest, and Support Vector Machines (SVM) (and general support vectors). A k-fold cross validation has been used to ensure that all possible combinations have been taken into account. To produce the best model output possible, Grid Search CV[ was also employed. The outcomes were then shown in a table to show the most effective and trustworthy techniques for evaluating loan requests.