Enhancing Credit Risk Assessment Using Bagging of Machine Learning Models

In today's highly competitive world, one of the major challenges for businesses is increasing sales. To achieve this, businesses adopt various strategies. One such strategy is offering credit payments. Credit payments are designed to alleviate the constraints customers face with immediate and cash payments. Not only does this method facilitate payment for customers, but it also increases the risk for businesses, as they face the possibility of non-repayment by the customer. This issue can be addressed using data-driven solutions and machine-learning techniques. Machine learning models can be trained using customer information, analyzing whether they have fully repaid their loans or defaulted, and subsequently predict whether a new customer will repay their credit. Essentially, this problem becomes a binary classification task. In this study, the Lending Club dataset is examined, and a pipeline for data cleaning and feature engineering based on a filter and, secondly, a wrapper method that uses a Random Forest classifier is proposed. Bagging models are then trained and tested using under-sampling and over-sampling techniques, and the performance of these models is compared with conventional models in the field of credit risk evaluation. It is observed that bagging machine learning models yield over ۹۳% AUC score compared to ۷۱% AUC score when these models are used individually.

Keywords:

Credit score prediction , Bagging , Ensemble learning , Supervised learning , Machine learning , Sampling techniques