CatBoost model and artificial intelligence techniques for corporate failure prediction

Financial distress prediction provides an effective warning system for banks and investors to correctly guide decisions on granting credit. Ensemble methods have demonstrated their performance in corporate failure prediction. Among the ensemble methods, gradient boosting has been successfully used in bankruptcy prediction. In this paper, we propose a novel approach to classify categorical data using gradient boosting decision trees, namely, CatBoost. First, we investigate the importance of the features identified by the CatBoost model. Second, we compare our approach with eight reference machine learning models at one, two and three years before failure. Our model demonstrates an effective improvement in the power of classification performance compared with other advanced approaches.

Automated summary

Financial distress prediction is crucial for banks and investors to guide credit decisions. The study introduces a novel approach CatBoost, using gradient boosting decision trees to classify categorical data, showing effective improvement in classification performance compared with other advanced approaches one to three years before failure.