CBR-Based Fuzzy Support Vector Machine for Financial Distress Prediction

摘要

Financial distress prediction has attracted wide attention since the 1960s. With the development of quantitative methods and computing tools, the models of financial distress prediction are constantly innovated. Commonly used models in this area include multiple discriminant analysis (MDA), logit, decision trees (DT), neural networks (NN), support vector machine (SVM), and case-based reasoning (CBR), etc. Support vector machine (SVM), because of its excellent generalization ability, has been a hot subject nowadays. Especially, fuzzy SVM (FSVM) has achieved great development in recent years. As choosing an appropriate fuzzy membership is an important issue in FSVM, in this paper we propose a new fuzzy membership for FSVM combined with CBR. Generally, our basic idea is to detect the outliers by their k-nearest neighbors, all or the majority of which are enclosed with the other class, and then assign them a lower fuzzy membership by the output of CBR. By adopting the hold-out method 30 times to generate 30 hold-out data, the empirical experiment shows the feasibility and validity of the proposed CBR-based fuzzy SVM for Chinese listed companies' financial distress prediction.