<![CDATA[The rapid advancement of digital technologies has encouraged the banking sector to adopt Artificial Intelligence (AI)-based approaches for credit risk management. Traditional credit scoring methods often lack accuracy in identifying default risks, particularly for unbanked and underbanked groups, leading to higher Non-Performing Loan (NPL) rates. This research addresses the need for a more adaptive, accurate, and inclusive credit risk assessment system in the digital economy era. This research aims to develop and evaluate an AI-driven predictive analytics model for credit risk assessment by comparing the performance of machine learning algorithms, such as Logistic Regression, Random Forest, XGBoost, and Deep Learning. The dataset comprises customer demographics (such as age and income), details of their banking relationship (including mortgage and securities account), and their response to the most recent personal loan campaign. The comparative analysis indicates that Random Forest substantially outperformed the other models, demonstrating superior accuracy (98.80%) alongside balanced precision (93.75%) and recall (93.75%), as well as the highest ROC-AUC (99.86%). These results highlight its robustness in both classification performance and discriminatory power. XGBoost and Deep Learning followed, showing competitive but lower predictive capabilities. In contrast, Logistic Regression exhibited clear limitations, yielding the lowest accuracy (90.40%) and precision (50%), despite achieving a relatively high recall (92.71%) and ROC-AUC (96.77%). This suggests that while Logistic Regression can identify positive cases, its overall reliability and precision are insufficient compared to advanced ensemble and deep learning methods.]]>
Copyrights © 2025
