Study On The Credit Score Model Based On The Selective Integration Algorithm

With the development of small-sum consumer credit and internet finance,the P2P network loan industry has developed rapidly and reached the trillion market.Individuals can register on the online loan platform and fill in relevant personal information,and can complete thousands to tens of thousands of dollars.Credit,and no mortgage required.However,at present,the concept of personal credit is extremely weak,information asymmetry,and adverse selection caused by adverse selection are serious.Therefore,online credit evaluation is very important.However,the personal evaluation index system of online lending is not perfect,especially the Internet consumption and social information are not taken into account.In addition,the accuracy and stability of the evaluation model need to be improved.At present,there are many problems in the single model and the combined model based on mathematical statistics and artificial intelligence methods(neural networks,random forest,and SVM),including high false positive rate,instability,weak generalization ability,and space and cost required for calculation.In response to the above-mentioned issues,the text proposes a systematic optimization of the scoring model,focusing on the improvement of the index system and the optimization of the scoring model.First,first analyze and analyze the traditional personal credit evaluation index system,and initially build an index system that meets the characteristics of online loans.Then,based on the logistic regression method,calculate the WOE and IV values of the indicators,perform significant analysis,and determine the index system scientifically and reasonably.Second,build a model based on the selective integration method,first select the classifier with accurate accuracy,calculate the Q statistical difference quantity,determine the final base classifier,and then introduce the selective integration algorithm,a heterogeneous number of classifier sets,Select the optimal subset integration model,build a selective integration model based on the OO ranking method,and selectively integrate the model based on the FCM-CFP clustering method.Third,in the end,the experiment was conducted using the data from the 360 platform and Renren Credit,and the two models were compared from the three dimensions of accuracy,base classifier size,and calculation time.In addition,the two models were compared with the single model and the combined model.The test results show that the two models have their own advantages and disadvantages.The selective integration model has the best predictive performance,and the sorting model has a shorter computing time.The clustering method has better stability due to better differences.Compared with single model and combined model,the number of base classifiers is more,but the score accuracy,stability and generalization performance are higher.