Comparing SVR, Random Forest, and XGBoost in ESG-Integrated Credit Risk Modeling: Evidence from Kenyan Banks

This research paper presents a comprehensive comparison of three machine learning algorithms for credit risk assessment when integrated with Environmental, Social, and Governance (ESG) factors. The study uses real-world data from Kenyan banks over a decade (2015-2024) to determine which algorithm provides the most accurate predictions and which ESG factors are most influential in credit risk assessment. Purpose: This study compares the predictive performance of three machine learning algorithms--Support Vector Regression (SVR), Random Forest (RF), and XGBoost--for credit risk assessment incorporating Environmental, Social, and Governance (ESG) factors. Design/Methodology/Approach: Using a balanced panel dataset of 15 Kenyan commercial banks from 2015-2024 (150 observations), we evaluate model performance using Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), coefficient of determination (R-squared), and Mean Absolute Percentage Error (MAPE). Findings: XGBoost significantly outperforms both Random Forest and SVR, achieving near-perfect predictive accuracy (R-squared = 0.998, RMSE = 0.0013, MAPE = 1.05%). Random Forest ranks second (R-squared = 0.946), while SVR exhibits the lowest performance (R-squared = 0.912). Environmental factors demonstrate the strongest inverse relationship with default probability (r = -0.647). Practical Implications: Financial institutions should prioritize gradient boosting methods for ESG-integrated credit risk models. Environmental performance metrics warrant greater weight in credit assessment frameworks. Originality/Value: This study provides the first comprehensive comparison of SVR, Random Forest, and XGBoost specifically for ESG-integrated credit risk modeling in an emerging market banking context.