Research On Evaluation And Pre-warning Model Of Individual Credit Based On Big Data

With the close combination of the financial industry and the internet industry,domestic financial institutions have regarded personal consumer loan business as one of their future development strategies,and personal credit risk control plays a crucial role in this region.In this paper,the user's behavior at all levels are digitized and integrated,and the user's consumption and behavior information are transformed into the assessment basis of personal credit by using parallelism and deep mining.The result can be used as individual credit risk control for financial institutions.At the same time,the efficiency and accuracy of traditional risk models show that they are not suitable for solving this problem.Therefore,how to combine the traditional risk assessment model and machine learning technology to obtain a more accurate assessment model based on the characteristics of variables in big data is a worthwhile study.The research contents of this paper focus on this issue.The main work includes:(1)Introduce and analyze the basic information of modeling such as data foundation,performance definition,sample classification and sampling scheme,and put forward the personal credit risk assessment model based on big data —— CEvaluation+.(2)Construct credit images of users from different dimensions,and elaborate the preprocessing methods such as data collection,data checking,data cleaning and factor analysis.(3)Constructing the evaluation and early warning model by combining efficient and interpretable Logistic algorithm with high accuracy and low data requirements Depth Learning model.The specific process is as follows: Firstly,the data to be processed using Logistic algorithm to predict the outcome of the bad users and intermediate users of the case.Then use composition model,decision trees and optimized genetic neural network model to generate a more accurate risk evaluation model.Moreover,assign the weights for the composite model according to the error rate of each model.Finally,the paper verifies the effectiveness of the CEvaluation+ model from the aspect of the ability of distinguishing and stability.It compares the model with other basic models and in order to demonstrate the effectiveness of the model,it compares to the scorecard model already in commercial use.Experiments show that efficiency and accuracy all have been improved in application of the CEvaluation+ model.