Class-imbalanced financial distress prediction with machine learning: Incorporating financial, management, textual, and social responsibility features into index system

Abstract

Financial distress prediction (FDP) is centrally imported to reduce potential losses of companies and investors. This paper combines social responsibility indicators with financial, management, and textual indicators to construct a multi-dimensional FDP index system. To increase prediction accuracy, the difference in the number of samples between special treatment and health companies is actively considered, and the synthetic minority oversampling technique is adopted to deal with class-imbalanced datasets. Moreover, a feature extraction method based on the genetic algorithm is employed to select features as input of machine learning models for predicting financial distress in Chinese listed companies. Experimental results show that ensemble classifiers are the better choice for predicting financial distress, of which gradient boosting decision tree outperforms other classifiers. Textual indicators play the most significant role in complementing traditional financial indicators of FDP, and their financial distress signals emerge earlier compared with management and social responsibility indicators.