Innovative Insights into Knowledge-Driven Financial Distress Prediction: a Comprehensive XAI Approach

In an era where the knowledge economy plays a pivotal role, corporate financial distress poses substantial risks to various external stakeholders. Traditional financial distress prediction models often fall short in accuracy and transparency. Hence, we introduce a groundbreaking approach that leverages the power of ensemble learning models while addressing the challenges of class imbalance. We propose a comprehensive framework that enhances predictive performance and caters to external users’ needs for interpretability. We delve into the world of explainable artificial intelligence (XAI) by incorporating techniques such as partial dependence plots (PDP), individual conditional expectations (ICE), and Shapley additive explanations (SHAP). These techniques provide global and local interpretations that help businesses, financial institutions, and regulators make informed decisions. We conduct empirical experiments using financial data from Polish companies, which show that the GBoost model with random oversampling (ROS) performs the best and demonstrates the validity of our approach. Local PDP interpretations help to identify key financial indicators that contribute to a firm’s financial distress, and ICE interpretations are exploited to provide strategies for improvement. Moreover, in terms of global interpretation, SHAP’s feature importance ranking and feature interaction results are consistent with the knowledge of financial experts, thus increasing the credibility of the black-box model. While our study presents significant contributions to financial distress prediction and XAI, it acknowledges certain limitations. Future research can explore heterogeneous ensemble predictors and extend this framework to various domains beyond bankruptcy prediction. Our research underscores the importance of knowledge-driven innovation in today’s dynamic economy.