Research on Credit Portfolio Optimization Decisions under Digital Risk Control: An Integrated Framework Based on Explainable AI and Hybrid Intelligence

To address the challenges faced by commercial bank credit management in the digital economy era, such as "black-box" models, subjective evaluations, and decision-making limitations, and to enhance the resource allocation efficiency of financial services for the real economy, this study constructs a three-stage integrated credit decision-making framework. This framework, drawing upon theories from explainable artificial intelligence (XAI), multi-criteria decision-making (MCDM), and meta-heuristic optimization algorithms, integrates "interpretable prediction, objective quantification, and global optimization." The framework utilizes an XGBoost-SHAP model to achieve transparent risk prediction and combines the Entropy Weight Method (EWM) with TOPSIS for the objective quantification of enterprise strength. A hybrid Simulated Annealing-Neighborhood Algorithm (SA-NA) is proposed to solve the dual-objective "return-risk" optimization problem for credit portfolios. The results demonstrate that the framework possesses acceptable predictive performance (AUC = 0.851) and interpretability. The optimized credit portfolio achieves an approximately 40% increase in RAROC (from 12.5% to 17.5%) and an 18.2% reduction in 95% CVaR (from 3.10M to 2.53M) compared to the rules-based baseline strategy. The proposed SA-NA algorithm significantly outperforms mainstream multi-objective optimization algorithms, such as NSGA-II, in terms of solution quality and diversity. This offers a potentially useful integrated framework that may support more informed credit resource allocation decisions.