Research On Personal Credit Scoring Sample Set Optimization Based On Monte Carlo Model

With the rapid development of commercial Banks, credit business is becomingincreasingly significant in the position in its developing process, the level of credit riskcontrol directly affect the progress of the whole commercial Banks. Therefore, thepersonal credit risk identification and evaluation of accurate or not, has become acommercial that bank can control the critical factors of risk. Past studies have focusedon the improvement of model accuracy, and ignores the fact that "reject inference"-sample bias that is an important problem. Because a lot of rating agencies can onlyaccept the part of the sample data through the credit model to predict loan applicantsdefault or not, resulting in sample bias problem will affect the validity of the creditrating model. Therefore, within the category of credit scoring, sampling bias is a need todeal with problems.This paper first analyses the sample bias and correcting technologies. Proposed therandom number generated by monte carlo method to produce sample added to theoriginal sample in the formation of new sample set so as to optimize the sample set,using the generated samples after optimization of sample set to solve the question ofsample bias. In produce samples, this paper expounds the generation method to generatesamples and choice, and to build personal credit scoring index system, determine thecommercial bank personal credit scoring index, based on commercial Banks index wereanalyzed, and the original sample to determine its distribution and the inner link,producing index data, classifying data corresponding repayment condition at the sametime, using software training form the sample. And add to the original sample qualitycustomers to generate sample rate and the collocation of generated sample and originalsample has carried on the structure and description, eventually forming optimization ofsample set. Finally, using the optimized sample set to field of single model andcombined forecasting model for the effect of test. Test results show that the optimizationof sample set as the sample data of personal credit scoring model can not only make itsforecasting precision is improved obviously, at the same time also can make sampleSelection Bias to obtain the very good solution.