The Applications Of Different States Of Ensemble Learning To Personal Credit Evaluation

With the rapid growth of Chinese economy and the widespread use of information and computer science in biological, economic and other fields,the demand for consumer credit of residents in all walks are increasing rapidly. Many major domestic banks are actively adapt to the market, it not only requires banks to extend diversified credit business for consumers in all levels, but also need banks to set a standardize system to evaluate consumer credit. This system can increase the banks'efficiency and low the risk in financial management. Therefore, the establishment of personal credit score system in China is so important for the development and stability in economic and it also can promote the prosperity of the financial market. Therefore, develop such a set of personal credit scoring methods can reduce the risk, not only has high academic value, but also has a strong use of significance.This issue of personal credit score, based on domestic and foreign personal credit score models and the principle of decision tree and ensemble learning approach, propose optimization the weight of a single model through using Adaboost and Bagging program. Firstly, starting from the principle of Adaboost, it can increases a variety of single classifier in the training process and achieve the minimum error rate which set in advance by adjusting the corresponding weight of weak classifier. Then in order to reduce misclassification rate, and improve the prediction accuracy. Analyze the whole process of removing noise by using Bagging ensemble to treat the unbiased estimate equally. Finally, the ensemble learning model has been trained by using the personal credit data. It showed ensemble learning methods optimize a single REP Tree in accuracy and robustness. The ensemble learning model combines the advantages of a single model and improve the credit scoring models'"over-fitting" and "local optimum" dilemma. The improved model which based on Adaboost and Bagging Methods has more advantages to reach the desired results of model.