Authors: Assistant Professor Dr.P.R.Venugopal, Assistant Professor Dr.Sreemathi Raghunandan, Assistant Professor Mrs.C. Sreedevi
Abstract: Corporate financial difficulties and defaults are costly occurrences that affect the firm, its investors, and the economy as a whole. In this paper, we design the Hybrid Predictive Analytics Framework (HPAF) to predict corporate financial risks by leveraging three different machine learning techniques, specifically, XGBoost with SHAP (SHapley Additive exPlanations) for credit scoring, Temporal Convolutional Network (TCN) with attention to model time sequence in quarterly financial statements, and Bayesian structural time series for macroeconomic sensitivities. Applying the HPAF to our benchmarking sample of 5,820 US public firms (from 2015 to 2025) which includes 412 firm-year default cases, HPAF scores higher AUC (0.932), lower Type I errors (12.4%), and lower Type II errors (9.7%) than Altman Z-score (AUC = 0.781), Merton structure models (AUC = 0.802), and even XGBoost alone (AUC = 0.871). On average, we can give warnings six quarters in advance before default occurs.
