The Application Of Ensemble Quadratic Surface Support Vector Machine In Personal Credit Risk Assessment

With the rapid development of the economy and the continuous innovation of Internet applications,the Internet finance industry has developed rapidly and received great attention from the government and society.However,the rapid development of Internet finance is also accompanied by the lack of risk control capabilities.The borrower's credit risk assessment and the establishment of a sound Internet financial risk control system have become a topic of concern for the further development of Internet finance.Because Internet finance companies have the characteristics of small amount,no mortgage,no guarantee,and poor customer qualifications,the traditional risk control method that relies on credit report of People's Bank of China is no longer applicable to Internet finance companies,but Internet finance company can mine Internet Big Data and use statistical and machine learning techniques to establish credit risk assessment models to solve this problem.Essentially,the personal credit risk assessment is a classification problem,so establishing a high-performance classification model is the key to solve this problem.As an extension of the support vector machine model,the quadratic support vector machine has the characteristics of better classification performance without using kernel function.The Ensemble learning achieves performance improvement by combining a set of base learners,and is usually more accurate than a single learner,and has already achieved great success in many real-world tasks.Therefore,this paper establishes ensemble quadratic support vector machine models and applies them to personal credit risk assessment.In the modeling process,firstly,the learning problem of the quadratic support vector machine is transformed into the Hinge loss function,which can be solved by the BFGS algorithm.Secondly,the Logistic function is used to transform the quadratic support vector machine decision values into posterior probability.Finally,because the quadratic support vector machine model does not consider the samples weight,this paper also establishes a weighted quadratic surface support vector machine model.After that,this paper establishes ensemble quadratic surface support vector machine models based on Real Adaboost,Discrete AdaBoost and Bagging methods,and carries out numerical experiments on three different data sets.The accuracy is used as the evaluation criterion.The results show that RAdaQSSVM has the best classification performance,and the DAdaQSSVM is slightly worse.However,since the QSSVM has better generalization ability,it is difficult for the BaggingQSSVM to construct the differences among the base learners,which results in lower performance improvement.Finally,the application of ensemble quadratic support vector machine in A Internet consumer finance company is introduced.Before modeling,some methods of data preprocessing and dimensionality reduction are introduced in combination with the desensitization data set provided by Company A.Using KS,AUC to evaluate the performance of the models.The experimental results show that the RAdaQSSVM model has the best classification performance.Based on the RAdaQSSVM model,a risk control process of Company A is proposed.Based on this process,the automatic approval rate of Company A reached 52.5%,and the rate of bad customers was only 1.98%,which saved manpower of Company A and reduced the company's bad customer rate.