Research On Personal Credit Evaluation Based On Portfolio Model

The development of personal consumption credit is not only conducive to guiding the planned consumption of individuals,but also can promote the healthy,stable and sustainable development of the national economy.How to promote the development of personal consumption credit while effectively managing the risks it brings is the key problem facing the consumer credit industry and related subjects.The common practice is to carry on the scientific reasonable forecast appraisal to the individual credit.It is a key problem faced by the theoretical and practical circles to predict and evaluate individual credit scientifically and reasonably and to develop personalized financial services on this basis.Studied in this paper based on the modern risk management theory,a bank of personal consumer credit behavior between the research object,the characteristics of the engineering,neural network,random forests,gradient tree such as machine learning algorithm,combination model for the technical methods,explore the scientific and reasonable method of personal credit assessment,and then to the consumer credit industry to provide scientific basis for the decision-making and related subjects.The results show that the application of feature engineering and combination model can not only overcome the defects of traditional logistic regression model,but also improve the prediction accuracy and performance of the model to some extent.The main contents of this research include: the first chapter is the introduction,which mainly introduces the research background and significance,domestic and foreign research status,research methods,ideas and contents of this paper;Chapter two describes the steps and related principles of characteristic engineering.The third chapter introduces the basic theories of logistic regression model,GBDT model,random forest model,neural network model,basic ideas and methods of combination model,and common indexes of model evaluation and comparative analysis.Chapter four is the main part of the research.Firstly,the original features of the data set of individual consumer credit behavior of a bank are processed by feature engineering,and 21 features affecting individual credit risk are selected.Data set in accordance with the ratio and then divided into training data and test data sets,and in the training data set will be the classic logistic regression model combined with neural network,random forests,GBDT,set up a combined forecasting model based on logistic regression and logistic regression model of a single,finally,and evaluation index of two model to evaluate the above four models,on the basis of comparative analysis,to explore the expression of the optimal model of data set,to predict the probability of default of the individual,and then to evaluate personal credit risk.The research results show thatthe four models have good prediction ability and risk discrimination,and the gbdt-logistic regression combination model established based on the research data set is superior to other models in terms of discrimination and prediction ability.The fifth chapter summarizes the whole paper and puts forward the problems and directions of the research.