Dynamic class-imbalanced financial distress prediction based on case-based reasoning integrated with time weighting and resampling

Jie Sun, Mingyang Sun, Mengru Zhao and Yingying Du

Abstract

Existing dynamic class-imbalanced financial distress prediction (FDP) models based on artificial intelligence, such as support vector machines or neural networks, are difficult to understand. Case-based reasoning (CBR) is an artificial intelligence method that is easy for users to understand, but traditional FDP models based on CBR lack mechanisms for treating concept drift and class imbalance. This study explores the construction of a dynamic class-imbalanced CBR FDP model, which consists of four modules (dynamic updates of the case base, class balancing of the case base by resampling, the time weighting of cases and CBR for FDP). It treats financial distress concept drift by dynamically updating the case base and via a time-weighting mechanism, and solves the class imbalance problem by resampling. Empirical experiments based on real-world data from Chinese listed companies show that the proposed dynamic class-imbalanced CBR FDP model outperforms both static and dynamic CBR FDP models without resampling or time weighting. Therefore, the dynamic class-imbalanced CBR FDP model not only gives a satisfying performance by effectively treating the problems of both financial distress concept drift and class imbalance but also has good interpretability in real-world applications, providing corporate managers and other stakeholders with a new risk management tool.