Research And Application Of Decision Tree Algorithm In The Classification Of Bank Personal Credit Users

As the rapid development of the economy, the living standard of people have improved a lot, coupled with the support of national policy, the personal loans have gradually been inseparable with people’s lives. To the banks, it’s a very important issue to divide the personal loan users and decide whether to agree the loan application. For the amount of personal loans is relatively small, not every loan application can be tracked which will greatly increase the cost. The data mining is considered to be a powerful technology to deal with such issues. With the help of personal credit evaluation model, we can greatly shorten the loan processing time, increase the accuracy of the prediction of the good or bad application, because the result all comes from the data, it is also a good way to reduce the influence of human factors.In this paper, we take the decision tree as the main modeling method. We build the personal credit score model to make a credit evaluation of loan applicants based on a true bank credit transaction data. With the help of the model, we can solve the high risk problem of personal credit risk. Through relevant experiments, the importance of each attribute were compared, then the accuracy of the model is contrasted, at last the feasibility of the dynamically generated algorithm is verified.The main work is listed as follows:(1) A comprehensive and systemic research and analysis on the general process of data mining and relevant algorithms is made by reading a lot of literature and document.(2) Take the German Credit Data the experimental data. Some data pre-processing work is made for modeling work. Then use the C5.0algorithm and the random forest algorithm to establish a personal credit evaluation model.(3) To better fullfill the need of bank personal credit business, we propose a random forest algorithm which generate the model dynamically and described the idea of algorithm and the implementation of the algorithm.(4) Through relevant experiments, the importance of each attribute in the personal credit data set were compared, then the accuracy of the model build though the C5.0algorithm and random forest is contrasted, at last the feasibility of the dynamically generated algorithm is verified.