A Multi-level Financial Distress Prediction Model Based on Rough Reduction and Clustering

Abstract

In order to improve the dynamic adaptability and predictive performance of the financial distress prediction model, this research proposed a multi-level financial distress prediction model based on rough reduction and clustering. This model improves predictive performance by the combination of an improved rough set attribute reduction method and the hierarchical clustering algorithm, BIRCH, which can process incremental data efficiently. Through attribute reduction by rough set, the influence of noisy data and redundant data were eliminated in order to identify the key indicators during the pre-processing phase. In the phase of FDP, the proposed multi-level model can deal with different application requirements so that different financial distress scenarios can be identified from various aspects. Empirical results with data from Chinese listed companies demonstrate that the model has a good dynamic adaptability and predictive performance.