Improved Support Vector Machine Personal Credit Rating Model

Abstract

Credit consumption has become the choice of more and more people. While the expansion of credit consumption brings convenience, the consequent problem of discredit brings incalculable loss to trust institutions, such as intentional arrears, malicious overdraft consumption, etc. to the trust in the operation of the huge losses caused. On the one hand, our country is unable to establish perfect personal credit record, customer credit information is difficult to share, access to relevant data is limited, personal credit system is not perfect, therefore, it is necessary to establish a perfect and automatic personal credit evaluation system, identify the personal credit risk scientifically, and realize the maximization of bank credit income. In the bank risk management, the main problem is how to measure and avoid the bank's financial risk, and personal credit assessment is the most difficult and important part. In this paper, we use the XGBOOST algorithm to sort the importance of features, and select the number of features and the specific features according to the relationship between the model accuracy and the number of features, then 80% of the data was taken into the traditional Support vector machine model as a training set and tested with the remaining data with low accuracy, so that, in terms of individual credit risk assessment, this paper introduces the quadric surface into the traditional Support vector machine and builds the kernel-free quadric surface Support vector machine model, which improves the accuracy by 9.8 percentage points and has a guiding significance for the bank reference risk control.