MACHINE LEARNING FOR ENHANCED CREDIT SCORING

Abstract

This study investigates the application of machine learning to improve the accuracy, fairness, and interpretability of credit scoring systems. Using the publicly available “Give Me Some Credit” dataset and following the methodology proposed by Ichim and Issa (2025), the research demonstrates how preprocessing techniques and advanced ensemble models, particularly Gradient Boosting Machines (GBM), enhance predictive performance. Key processes included data cleaning, class balancing with SMOTE, and feature scaling, which collectively improved the model’s Area Under the Curve (AUC) from 0.83 to 0.87. Figures derived from the model illustrate the most influential features, compare model discrimination through ROC analysis, and highlight the impact of balancing strategies on performance. The study also emphasizes the importance of model interpretability and regulatory transparency in the adoption of machine learning within financial decision-making. This research contributes a replicable, interpretable framework for institutions seeking to modernize credit risk assessment while maintaining compliance and trust.