Personal Credit Scoring Modeling Based On Data Mining

With the rapid development and the objective law of risks and return of consumption credit business, the financial institutions hunt huge profit; meanwhile it have to face with tremendous credit risks. Therefore, how to avoid credit risk is an important subject to banks and crediting organizations. The banks need to make scientifically credit evaluation of information of credit costumer before granting credit loans, objectively, comprehensively and accusatively evaluate repayment capability and repayment willing of consumer, and avoid, control and reduce bad debt loss. In the western countries, it is commonly adopted personal credit evaluation of quantitative analysis to evaluate credit status of individual customers. By means of data mining technology, banks and crediting organizations can establish credit scoring models and dig useful model and regulation by deeply analyzing data information of applicants, as the decision basis for credit management. At present, because of the imperfect social credit reporting system and backward credit consumption industry, commercial banks exploiting personal credit models just start, and lack experience in exploiting suitable models. This article will study this.Credit scoring technology is usually divided into statistical method and non-statistical method. Statistical method includes linear regression, discriminate analysis, logistic regression, and decision tree etc. Non-statistical method includes linear programming, neural network and genetic algorithm etc. But now, which method is best one to development technology of credit models has no consistent conclusion. In this paper, it is used most common scoring technology- discriminate analysis, logistic regression and neural network to study on adaptability using as analysis object with credit data. Utilizing them to build models respectively, classify the customers and compare analysis result. The results showed that the three models have some prediction ability and distinguish between good costumes and bad costumes, but logistics regression mode is the best one in the three scoring technologies. It is the optimal model that commercial banks can adopt nowadays and is worth popularizing in practical.In the process of building credit models, preparing data is very important. Preparing data is the foundation of building personal credit scoring model. But the data of collection in practical was polluted. So it is necessary to preprocess for data before building credit model. In this paper, it studied on data preprocessing including data cleaning, data conversion and data clustery, especially categorical variable with more characteristic item in data.Finally, the process of building models are concluded, suggestions for the need of practical application are put forward. Existing some disadvantages in the paper and subject require deep research in the further are pointed out.