Research On Personal Credit Evaluation Model And Algorithms Based On The Support Vector Machines Thesis

At present, personal consumer credit has been an important measure to spur our country's domestic demand and promoted the development of economic. In the course of individual's consumer credit development, one of the main problems we faced is the difficulty in evaluating and controlling personal credit risks which cause the high risk of credit consumer business. Therefore, the research of the personal credit indicator system and evaluation algorithm is becoming more and more important.The paper takes account of three primary projects of personal credit evaluation system model, data preparation and evaluation algorithms. The achievements in the research are as follows:Firstly, a 7-tuples overall framework of the personal credit evaluation system is put forward to solve the problem that the system is lack of the description of the formalization and the standardization. Moreover, a personal credit evaluation system module based on the SVM thesis is built, which points out the key issues necessary for the system implementation.Secondly, an evaluation indicator system constituted by 10 indicators such as sex, age and income is brought out, depending on the construction regulations of evaluation indicators authorized by the industry and the academics, the influence to the personal credit evaluation by respective indicator and the references of the indicator system supposed by the Chinese banks.In order to improve the effect of the personal credit evaluation, the paper investigates the data pro-processing issue before the evaluation and provides the pro-processing method of the personal credit data such as data cleaning, data integration, data presentation and data transformation.Finally, the paper brings forward a BTFSVM algorithm to meet the requirement of the multi-classification and the anti-inference of evaluation algorithm, a BTFSVM algorithm supporting the incremental learning on the basis of the SISVM algorithm and the BTFSVM algorithm to meet the requirement of the incremental learning of the evaluation algorithm, and a BTFSVM algorithm supporting the parallel learning on the basis of the Cascade SVM algorithm and the BTFSVM algorithm to meet the requirement of the parallel learning of evaluation algorithm. Meanwhile, the research proves the availability and the feasibility of the three algorithms by the corresponding experiments.