With the rapid development of China’s commercial banking, personal creditbusinesses play an increasingly important role. The effects of personal credit scoringdirectly affect the bank’s business development level and the bank’s overallrequirements. Therefore, the level of banks control risk depends on the accuracy ofidentification and scoring of personal credit.This thesis summarizes the research status of traditional scoring models at thebeginning. The traditional scoring models mostly based on statistical methods andartificial intelligence methods, which have some problems such as instability and poorpredictive accuracy. These problems seriously hampered the development of personalcredit evaluation. This thesis starts the research from the personal credit dates, analysisof the characteristics of the personal credit dates and requirements of credit data forpersonal credit scoring model, conducts data mining from sample optimizationperspective, achieves sample dimensionality reduction and de-noising, utilizes multipleIndicator stratified sampling method to filter and extract sample data in order to avoidsample bias problem, establishes sample set of good quality properties. Moveover,thisthesis uses Bayesian network to build personal credit scoring model based on optimizedsample set, Bayesian network can perform learning and reasoning in the limited,incomplete and uncertain information conditions. The results can be well explained byprobability theory and scoring model is very stable by stabilized network structurebased on prior probability. The results of the Bayesian network based on sample setoptimization indicate that it can effectively improve the prediction accuracy of scoringmodel by using sample set optimization, at the same time, the stability of the scoringmodel is keeping on high level. |