Integrating Financial and Non-Financial Indicators through RF-SVM-Stacking Model for Accurate Green Credit Risk Assessment

Abstract

Green credit has emerged as a crucial financial mechanism to promote sustainable economic development and mitigate environmental degradation. However, the evaluation and risk assessment of green credit remain a significant challenge due to the complexity of environmental factors and the limitations of traditional financial scoring models, which primarily rely on quantitative financial data. This study aims to develop a more accurate and comprehensive green credit scoring approach by integrating financial and non-financial indicators into an advanced hybrid model. To achieve this, an RF-SVM-Stacking integrated model is proposed, combining Random Forest (RF) for feature importance ranking and Support Vector Machine (SVM) for credit scoring. The model incorporates conventional financial indicators along with non-financial factors, including green credit risk characteristics, innovation input indicators, and ESG (Environmental, Social, and Governance) ratings. Methodologically, the stacking ensemble technique is employed to enhance prediction accuracy and robustness across datasets. The empirical analysis demonstrates that the proposed RF-SVM-Stacking model achieves higher accuracy and better generalization capability compared to baseline models such as SVM with Bagging or AdaBoost, neural networks, and Gradient Boosted Decision Trees (GBDT). The findings suggest that incorporating non-financial and sustainability-related metrics significantly enhances the accuracy of green credit risk assessment. These results have important implications for financial institutions and policymakers, suggesting that adopting integrated machine learning approaches can effectively support the development of a sustainable financial system and guide more responsible investment practices aligned with global environmental objectives.