Research On Individual Credit Evaluation Based On Multi-model Combination

With the rise of the Internet financial industry,the credit consumption of Chinese residents has developed rapidly,and the total personal credit has increased year by year.As of the end of 2019,the non-performing loan balance of personal credit reached 2.41 trillion yuan.Doing a good job of personal credit evaluation is especially critical.In practical applications,continuously improving the accuracy of evaluation models is the main research direction of personal credit evaluation.Ignoring the differences in the importance of data characteristics will reduce the accuracy of the model to a certain extent;in addition,the interpretability of the model needs to be taken into account to enable funders and customers to understand credit decision rules.The classification accuracy of deep neural networks is high,but it is limited by the interpretability of the model,and is rarely used in the practical application of personal credit evaluation.The accuracy of deep forest algorithms and deep neural networks are indistinguishable from Xuanyuan,and their interpretability is better.The multi-model combination method can complement the advantages between models.Therefore,based on the multi-model combination method,this paper selects representative deep forest,LightGBM,and SVM models to build a personal credit evaluation model.The main work of this paper is as follows:(1)Improve the classification algorithm.Improve classification algorithm.During feature extraction,the features are partitioned according to their importance,and then feature subspaces are constructed from different partitions in proportion to the features.Assign weights to decision trees in deep forests,and propose improved deep forest algorithms to improve classification accuracy.(2)Verify the multi-model combination method.First,considering that the number of models increases and the parameters also increase,an improved grid search algorithm is proposed to improve the efficiency.The range of the optimal parameter points is searched with a large distance,and the optimal parameters are determined with a small grid distance.Then,the multi-model combination method can be used to complement the advantages of the models,select a deep forest model with good stability and accuracy,and combine the LightGBM and SVM models quickly and efficiently,and compare the effects of multiple combination methods through experiments.(3)Empirical analysis and comparison.The combination model with the best experimental results is compared with three single models,and the experimental results are analyzed to determine the final selection plan.Finally,the selected combination model is empirically compared with several commonly used models to verify the performance of the combination model.Experiments through the public personal credit data set of the "Lending Club" company,the results show that our algorithm are compareble to others.