Financial distress warning and risk path analysis for Chinese listed companies: An interpretable machine learning approach

Financial distress is typically not a sudden occurrence, but rather the outcome of accumulated operational inefficiencies and external pressures. In China’s capital market, existing financial distress warning models offer limited interpretability, making it challenging for regulators to obtain a reliable basis for risk identification. To address this limitation, we propose an interpretable machine learning framework that integrates extreme gradient boosting with non-dominated sorting genetic algorithm II for multiobjective optimization, and Shapley additive explanations with interpretive structural modeling to reveal both the marginal effects and the risk formation pathways of financial indicators. Using empirical data from A-share listed firms between 2010 and 2024, the optimized model demonstrates a 3.32 % improvement in warning accuracy and a 2.15 % gain in efficiency compared with benchmark models. Furthermore, the findings show that the predictive influence of profitability diminishes as the lead time before financial distress increases. Overall, this study presents an interpretable model that enables regulators and policymakers to identify financial risks at earlier stages and implement targeted interventions in the market environment.