An Explainable Artificial Intelligence Framework for Intelligent Financial Analytics, Accounting Automation, and Strategic Corporate Decision-Making

Financial distress prediction for corporations demands analytical systems that not only discriminate well, but also are sufficiently transparent to allow for responsible management actions. This work proposes an explainable AI system to automate the analysis of accounting-based financial indicators to support strategic bankruptcy-risk decision-making. A leakage-resistant development – validation – locked-test design was used to analyze the Taiwanese Bankruptcy Prediction dataset of 6,819 firms and 95 financial variables. Stratified cross-validation was used for tuning Logistic Regression, Random Forest, Extra Trees and Histogram Gradient Boosting with precision-recall area under the curve (PR-AUC) serving as the main selection criterion due to the low proportion of bankrupt firms (3.23% of observations). Extra Trees had the perfect validation profile with a locked-test PR-AUC of 0.4567, ROC-AUC of 0.9422 and Brier of 0.0227. Validation based risk bands focused 87.88% of bankruptcies in the locked test in the High and Critical categories and the permutation importance identified borrowing dependency, interest-bearing debt cost and financial leverage. It shows how explainable and calibrated financial-risk analytics can be used for an automated accounting-based screening process while maintaining interpretable evidence for corporate decision making.