Financial Distress Prediction of Foreign Trade Listed Companies in China Based on the Hybrid Model of Lasso-SMOTE-XGBoost-SHAP

HRMARS - Persistent turbulence in the global trade product severe uncertainties in the operation and financial sustainability of listed foreign trade enterprises. To accurately identify and forecast the financial distress and suggest risk management support for corporate decision-making, this paper takes Chinese A-share listed foreign trade firms spanning 2005 to 2025 as research objects, constructs a multi-dimensional indicator system combining financial and non-financial metrics, and further develops a hybrid Lasso-SMOTE-XGBoost-SHAP model for financial distress prediction. Specifically, Least absolute shrinkage and selection operator (Lasso) regression is used to multi-dimensional feature screening, the synthetic minority oversampling technique (SMOTE) solve the sample class imbalance problems, eXtreme gradient boosting (XGBoost) realizes binary distress classification; and the shapley additive exPlanations (SHAP) quantifies each feature’s marginal contribution and influence direction. The results suggest that comparative experiments verify the outstanding performance of the proposed integrated model. When trained on the dataset that incorporates both financial and non-financial indicators, the model achieves precision of 0.9574, accuracy of 0.9617, recall of 0.9663, F1-score of 0.9618 and an AUC value of 0.9279, surpassing other four benchmark models. SHAP confirms that type of audit opinion, Debt-to-Tangible-Assets Ratio are core risk predictors, proving audit and governance information delivers substantial incremental early-warning capacity.