Multiple Classifiers Hybrid Combination For Companies' Financial Distress Prediction

Abstract

Abstract In order to control uncertainty and instability of single classifiers in financial distress prediction, this research proposed a multiple classifiers hybrid combination model for financial distress prediction. This model improves predictive performance by the combination of multiple classifiers and taking advantages of serial combination and parallel combination. Diversity principle and individual optimization principle were taken as criteria for classifier selection. On the foundation of defining selection operator for class's best classifier, algorithm for constructing basic modules in hybrid combination, dynamic weighting mechanism for parallel modules inside hybrid combination, and mechanism of majority voting were designed. Empirical research with data from Chinese listed companies indicates that the model improves average predictive accuracy and simultaneously reduces variation degree. Statistical analysis demonstrates that the hybrid combination model outperforms existing single classifiers in financial distress prediction significantly.