Research On Personal Credit Evaluation Based On Credit Platform Data

Credit rating is the basis for financial institutions to make credit decision.With the development of science and technology,big data technology has penetrated into the financial field,and personal credit has got into a new era.Personal credit evaluation based on big data is one of the hot spots in the current research.This paper conducts a comprehensive experimental study based on personal credit data.The characteristics of high dimension and low standard of personal credit data increase the difficulty of experiment.In order to get better sample data and reduce data dimension,this paper studies the feature selection algorithm such as random forest,Pearson correlation coefficient and distance correlation coefficient.On this basis,combined with the heat distribution map of feature correlation,it optimizes the correlation of data features,and creatively proposes the Pdc-RF algorithm.Furthermore,we tested the performance of Pdc-RF algorithm by simulation,which shows better performance compared with traditional feature selection algorithm.Specifically for the specific application scenario of personal credit data credit evaluation,the experimental data set is cleaned,the discrete data is one hot coded,and then all data features are standardized.Based on the Pdc-RF algorithm,the feature selection of personal credit data is completed,and the 145 dimension data is reduced to 22 dimension.On this basis,the data after dimensionality reduction were analyzed in the aspect of statistics.On the basis of data preprocessing and feature dimensionality reduction,this paper builds a score card model of personal credit evaluation model.In the process of modeling,first of all,data partition,woe coding transformation and IV value calculation are carried out for the experimental data.In order to obtain the best evaluation model,this paper based on the three machine learning algorithms of logical regression,random forest and SVM support vector machine for training.After the adjustment of super parameters and performance comparison,we choose the logic regression model with the highest AUC and K-S values.Finally,according to the prediction results of the output of the logistic regression model,this paper gives the credit user feature-based scorecard.