Empirical Analysis On Network Credit Personal Credit Evaluation

With the Internet,large data and other technologies rapidly developed in the country,the rise of a large number of Internet financial enterprises to use its technological advantages to provide customers with more convenient and efficient financial services,of which the network loan business focused on by the financial industry.At the same time,the default risk will be a major risk of Internet finance.How to effectively anticipate and identify potential default users,Internet finance has become a difficult problem in the risk management of the urgent need to overcome.In this paper,combined with large data and social networking technology,a personal credit assessment model is designed for a financial platform.This paper first introduces the Internet financial background,current situation and problems faced by Internet loan business,studies the research status and evaluation model of credit risk both at home and abroad,and emphatically elaborates the principles and algorithms of decision tree combination algorithm(including random forest,Adaboost and Gradient Boosting decision tree),social networking visualization technology etc.knowledge.Secondly,based on the realized transaction data provided by the platform called Rong360,the data were preprocessed by variable derivations and screening.The decision tree model(random forest,Adaboost and GBDT)and ten fold cross validation were used to train and compare.The accuracy of the three combinatorial models is obviously superior to that of the single decision tree,and the average accuracy of the GBDT is the highest.Besides,the grid search method is used to find the optimal parameter combination of the GBDT model.Finally,the GBDT model with interaction.depth 2 and n.trees 360 is taken as the personal credit evaluation model.In order to exploit the social network behavior on the platform,we construct a social network G(1983,1162)between users.Through the the visual network structure,most users of the network is relatively simple,by one-to-one relationship.But only a few influential users have a small social network.After further investigation of the connected components,it is found that the network is dominated by small connected subgraphs,and the correlation between users is not significant.The default rate of the connected components(the number of nodes is 8)is the highest,and the default rate is mainly concentrated in two extremes,namely,the number of nodes is 2,3 and 46,and the default rate of the appropriate size is low,that is,the level of default is not only connected with the number of nodes,but also with the user in the location of the network.Based on the above research results,the FR algorithm is used to lay out the social network of two types of users,and the internal structure is analyzed visually.Compared with normal users,the social relations of the default users are more complicated,and the groups are composed of different relationships.But in the normal user’s network,the relationship between nodes and nodes is relatively simple.Finally,the credit model is summarized and the future work is prospected.