Determining the credit worthiness of retail banking customer by machine learning technique

• Main Author: Peng, Yangling
• Other Authors: Wong Kin Shun, Terence
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2024
• Subjects: Engineering; Machine learning
• Online Access: https://hdl.handle.net/10356/176996
• Institution: Nanyang Technological University

This research project conducts a comparative analysis of statistical and machine learning models for credit risk assessment, focusing on their performance in the context of imbalanced datasets common in loan default prediction. Traditional statistical models and advanced machine learning algorithms, including Decision Trees, Support Vector Machines (SVMs), and K-Nearest Neighbors (KNN), were evaluated to determine the most effective approach for predicting creditworthiness. Machine learning models outperformed their statistical counterparts, adeptly identifying complex, non-linear patterns and adjusting to data imbalance. Resampling techniques, especially AdaptiveSMOTE, enhanced the predictive accuracy of distance-based models such as KNN and SVM by ensuring a balanced class distribution, crucial for minority class prediction. Statistical models faced challenges with the skewed original dataset, which were mitigated by implementing SMOTE and AdaptiveSMOTE to balance class representation. These resampling strategies proved critical in improving the models' prediction reliability. The study addresses the acute challenge in banking: the scarcity of default data. The improved analytical methods bolstered the banks' capacity for precise credit risk management, thereby reinforcing the financial sector's defenses against potential defaults.