Credit scoring with an improved fuzzy support vector machine based on grey incidence analysis

Credit scoring has become increasingly important as the economy recovers, and thus a huge amount of customer credit data is collected by commercial banks and finance corporations. With the rise of machine learning, credit risk can be assessed more easily according to historic data, and support vector machine (SVM) is considered to be an “off-the-shelf” supervised learning algorithm to solve the classification problem successfully. In this paper, an improved fuzzy support vector machine (FSVM) is proposed to overcome the classification problem caused by noise and outliers. First, the notion of mean grey incidence degree is defined to describe the relevance among the training samples. Then, homogeneous and heterogeneous class centers are selected as two reference points in order to discriminate noise and outliers from the valid data. Finally, a fuzzy membership function is given for the purpose of FSVM training. As an empirical study, two credit data set are chosen to demonstrate the feasibility of the model.