Credit risk assessment using polynomial logistic regression and ridge regularization

Abstract

Effective credit risk assessment is essential for financial institutions, particularly Savings and Credit Cooperative Societies (SACCOs), as it helps evaluate borrower creditworthiness and reduce loan defaults. Traditional predictive analytics methods, such as Naïve Bayes and Decision Trees, have been widely used for credit risk assessment. While these methods perform reasonably well, they often struggle to capture complex, non-linear patterns and imbalanced distributions in credit data, leading to suboptimal predictive accuracy and model stability. Additionally, these methods assume linearity and independence among predictors, which fails to accurately reflect actual borrower behavior. This study aims to develop a high-performance predictive model for credit risk assessment, using an enhanced Logistic Regression approach. The model incorporates polynomial transformations to capture non-linear interactions, Recursive Feature Elimination (RFE) to rank and select the most relevant predictors, and ridge regularization to handle complex feature relationships and ensure stability. The model was trained and evaluated using a real-world credit risk dataset from Kaggle, achieving 90% accuracy, 94% recall, and an AUC of 0.95. A comparison with XGBoost and Random Forest models revealed that the proposed model delivers a strong balance between accuracy and computational efficiency. This study shows that enhancing logistic regression with RFE, polynomial transformations, and ridge regularization significantly improves credit risk prediction while maintaining the interpretability crucial for financial institutions.