Explainable AI for Banking Stability Prediction: Integrating Quantum-Inspired Fragility Metrics with Random Forest and SHAP Analytics
Main article
Abstract
The opacity of machine learning models poses a critical challenge for their deployment in financial supervision, where regulators and analysts must understand and justify predictive decisions. This study presents an explainable artificial intelligence (XAI) framework for banking stability prediction that integrates quantum-inspired fragility metrics derived from double-well potential modelling with a Random Forest classifier interpreted through SHapley Additive exPlanations (SHAP). Using an annual panel of twelve Mexican commercial banks observed over 2014–2023, we construct a comprehensive feature set that combines classical bank-level financial indicators — including the non-performing loan ratio, capitalization index, net interest margin, and return on assets — with novel quantum-inspired predictors generated via functional structures of Quantum Field Theory (QFT). The proposed augmented Random Forest model achieves an AUC-ROC of 0.887 and an F1-score of 0.858, outperforming all baseline classifiers. SHAP analysis reveals that the quantum fragility score ranks as the third most influential predictor, behind non-performing loans and the capitalization index, confirming that physics-inspired indicators carry independent informational value beyond traditional financial ratios. SHAP dependence plots further demonstrate non-linear interaction effects between the capitalization ratio and the quantum fragility score, consistent with phase-transition dynamics identified in the QFT literature. The framework advances both the theoretical integration of quantum finance methods and the practical need for interpretable early-warning tools in emerging banking markets.
