Multi-Channel Prediction for Probability of Bank’s Customer Credit Default Based on Machine Learning Techniques along with an earlier Recommendation System رسالة ماجستير

Abstract

Lending institutions and banks are surrounded by a multi-risk business environment, the most important of which are those related to clients' default and non-fulfillment of their obligations towards receivables, which leads to their inability to continue their activities and then failure and bankruptcy. The main challenge is to identify the most important indicators and variables that lead to the customer's default credit and build a hypothetical forecasting model that can effectively and accurately predict the probability of default before it occurs along with the recommendation system (RMSS). This is essential to take the necessary decisions and measures that prevent customers from defaulting in the future and to have access to a credit portfolio that is fairly free from defaulted accounts. This study was conducted on data provided by the Jordan Ahli Bank in Palestine, for 8506 borrowers between regular customers and defaulting customers. The dataset is for this study exclusively and has been approved by the Bank's management. In this study, different machine learning algorithms have been implemented because the probability is an expected value, the accuracy value of the most used methods exceeds 90%, and SVM is superior to other classification models. we propose an RMS based on an item-based collaborative filtering technique (which uses the customer preference data from the algorithm) that works side by side with machine learning algorithms to provide the necessary recommendation to avoid the client getting default. Our findings illustrate important techniques for detecting and predicting credit failures before they occur. Our methodology and results will help regulators, banks, and creditors to meet the challenges of credit granting and financial catastrophe and maintain a clean credit portfolio, by addressing various financial risks and credit rating improvements.