An Enterprise Financial Health Rating System Based on Deep Belief Networks in Financial Big Data Analytics

The global textile and leather industry faces mounting pressure from market volatility, supply-chain uncertainty, financial risk, and ESG-related disclosure requirements. This study constructs an enterprise financial health rating system based on a Deep Belief Network (DBN). The sample includes 247 listed companies in the textile, apparel, and leather products sectors from the Shanghai-Shenzhen A-share markets and the Hong Kong Stock Exchange during 2013–2023, yielding 2,486 firm-year observations. Eighteen core features are extracted across five analytical dimensions: solvency, operational efficiency, profitability, cash flow quality, and ESG financial disclosure. A three-layer stacked Restricted Boltzmann Machine architecture combined with Bayesian hyperparameter optimization is used to perform seven-tier classification. The DBN model outperforms Logistic Regression, SVM, XGBoost, and LSTM across major metrics, achieving an AUC-ROC of 0.9218, a macro-averaged F1 score of 0.8312, and a KS statistic of 0.7134. Ablation experiments further show that incorporating ESG financial indicators improves high-risk rating identification accuracy by 3.24 percentage points. SHAP interpretability analysis provides traceable evidence for rating decisions. The system can therefore serve as a deployable intelligent rating tool for credit institutions and supply-chain finance platforms.