Predictive Business Analytics for Banking Fragility: Machine Learning and Quantum-Inspired Indicators in Emerging Financial Markets
Main article
Abstract
Banking fragility in emerging financial markets remains a persistent concern for prudential supervisors because traditional linear econometric models often fail to capture the non-linear and discontinuous dynamics characteristic of these systems. This study develops a predictive business analytics framework that combines machine learning algorithms with quantum-inspired functional indicators to assess the fragility of the Mexican banking sector during the 2014 to 2023 period. Using annual panel data covering ten domestic commercial banks, the framework integrates classical microfinancial ratios with two families of quantum-inspired predictors derived from double-well stochastic potentials and Faddeev–Popov restricted quantization. Four classification models — logistic regression, random forest, gradient boosting (XGBoost), and a quantum-enhanced random forest — are trained, cross-validated, and compared on a binary fragility outcome defined by the lower quartile of the bank Z-score. The quantum-enhanced random forest achieves an area under the ROC curve of 0.881, outperforming the logistic baseline by 17.7 percentage points and the standard random forest by 6.9 percentage points. Interpretability is assessed through SHAP-based feature attribution, which confirms that the quantum-inspired fragility index and the Faddeev–Popov ghost entropy carry the largest predictive weight, alongside non-performing loans and capitalization. The results demonstrate that quantum-inspired features capture latent risk dynamics that escape classical indicators, and that their integration into transparent machine learning pipelines offers a viable path for early-warning systems in emerging economies. The framework contributes to the data analytics literature by bridging theoretical physics, predictive modelling, and prudential supervision practice.
