基于模糊支持向量机的上市公司财务困境预测

Predicting financial distress of listed corporations based on fuzzy support vector machine

支持向量机(SVM)已经成功地应用于财务困境预测问题的研究,且已证明优于多元线性判别分析(MDA)、逻辑回归(Logistic regression)和神经网络(NN)等方法.然而,传统SVM使用结构风险最小化的原则,这样可能导致错误分类的经验风险升高,特别是当样本点与最优超平面十分接近的时候,这种误分类的经验风险显著升高.另外,传统SVM还存在过拟合问题,所以对数据集中的外点或噪声十分敏感.因此,采用模糊支持向量机(FSVM)算法来改进上述不足.首先,建立一个适当的成员模型用于对整个数据集的模糊处理;然后通过外点侦察方法(ODM)来发现外点,其中ODM集成了模糊C-均值算法(Fuzzy C-mean algorithm)和无监督神经网络中的自组织映射(SOM).最后,为主体集和外点集中的样本点分配不同的权值.还将FSVM应用于上市公司财务困境预测的实证研究,实证结果表明FSVM与传统SVM相比,FSVM能较好的解决经验风险升高和过度拟合问题,确实降低了外点的影响并提高了分类器的分类准确率.

The support vector machine （SVM） has been applied to the problem of bankruptcy prediction, and proved to be superior to competing methods such as the linear multiple discriminant approaches, logistic regression and the neural network. However, conventional SVM employs the structural risk minimization principle, thus empirical risk of misclassification may be higher, especially when a point to be classified is close to the optimization hyper-plane. Furthermore, SVMs are very sensitive to outliers or noises because of overfitting problem. In this paper, FSVM was proposed to deal with the corporate financial distress prediction. First, a proper membership model was used to fuzzy all the training data of positive/negative class. Second, the outliers were detected by the proposed outlier detection method （ODM）. The ODM was a hybrid method based on the fuzzy c-means （FCM） algorithm cascaded with an unsupervised neural network, called self-organizing map （SOM）. Finally, FSVM was applied to corporate financial distress prediction, and experical results indicated that the proposed FSVMs actually reduced the effect of outliers and yielded higher classification rate than SVMs.