Enterprise Financial Distress Prediction Based on Machine Learning and SHAPInterpretability Analysis

Accurately predicting corporate financial distress is a critical aspect of modern financial risk management, especially amid growing economic uncertainty. Traditional analytical methods often rely on limited financial indicators and expert judgment, which cannot capture complex, nonlinear patterns. To address these limitations, this study develops a comprehensive machine learning framework using a dataset comprising 3,672 samples and 83 financial features. Multiple predictive models were constructed and evaluated, with XGBoost demonstrating superior performance (F1-score = 0.242, ROC-AUC = 0.910). To enhance interpretability, SHAP (SHapley Additive exPlanations) was employed post hoc to identify influential features and provide transparent, actionable insights into model decision-making. The analytical pipeline includes data preprocessing, correlation-based feature selection, model training, and performance evaluation. SHAP analysis further highlights the significance of features such as x36, x44, and x13, underscoring the framework’s utility as a reliable tool for financial risk assessment.Beyond technical performance, the framework offers practical significance for small and medium-sized enterprises (SMEs), which often lack access to advanced risk forecasting tools due to resource constraints. By enhancing early warning capabilities and decision- making transparency, this cost-effective solution supports operational resilience, informed financial planning, and long-term economic sustainability.