The performance of corporate financial distress prediction models with features selection guided by domain knowledge and data mining approaches

Abstract

Experts in finance and accounting select feature subset for corporate financial distress prediction according to their professional understanding of the characteristics of the features, while researchers in data mining often believe that data alone can tell everything and they use various mining techniques to search the feature subset without considering the financial and accounting meanings of the features. This paper investigates the performance of different financial distress prediction models with features selection approaches based on domain knowledge or data mining techniques. The empirical results show that there is no significant difference between the best classification performance of models with features selection guided by data mining techniques and that by domain knowledge. However, the combination of domain knowledge and genetic algorithm based features selection method can outperform unique domain knowledge and unique data mining based features selection method on AUC performance.