Explainable machine learning for predicting zombie firms: a three-dimensional analytical framework

This study proposes an explainable machine learning model to predict zombie firms based on a three-dimensional analytical framework, which captures firm characteristics across internal value creation, external financial support, and prolonged financial distress. Using a panel dataset of 55,986 observations from 5,151 Chinese listed companies, we evaluate eight machine learning algorithms. Among them, LightGBM performs best, a finding that holds across a series of robustness checks. By combining SHAP with LightGBM, we find that internal value creation is the most explanatory dimension, with profitability showing the highest SHAP value. Further heterogeneity analysis reveals that the structure of predictive contributions varies across firm ownership, industry, and region. Our findings provide an explainable approach for the early detection of zombie firms. The study also offers practical insights for improving resource allocation and supporting industrial restructuring in emerging markets.