Weighted Support Vector Machines for Building Credit Scoring Model

Credit risk assessment has become an increasingly important area for financial institutions for recent financialcrisis and regulatory concern of Basel II. The quantitative credit scoring models have been developed to help creditmanager evaluate customers’ credit risk for decades of years. Most of the existed SVM models for credit scoring focus on improving the classification accuracy with giving the same penalty weight to classifying error for both the good and the bad. In practice, the credit manager need the flexibility of the model to adjust the weight. In this paper, we introduce 1-norm weighted support vector machines and weighted least square support vector machines to build a credit scoring model and test their performance with different measurements on a real-world datasets. The results show that weighted SVMs have good flexibility and different objectives can be satisfied by choosing proper parameters of the models.