Research On Personal Credit Risk Assessment Based On Machine Learning

With the improvement of the living standard of residents,the popularization of the concept of early consumption and the growing credit card market,the amount of credit card issuance has entered a period of rapid growth again.For commercial banks,it is both an opportunity and a challenge.The rapid increase of credit card issuance not only brings more Interest-bearing Capital and interest income,but also accompanies the positive growth of credit risk.More and More commercial banks focus on how to quickly and accurately assess the individual credit risk.Personal credit risk assessment is generally converted to a two-class problem,and various data mining classification models are widely used as personal credit risk prediction models.With the arrival of the era of big data,the traditional personal credit risk assessment model brings problems in training and prediction of the model under the circumstances of large amount of data,complex structure and increasing dimensions.Deep learning can use large data for model training.Compared with constructing features by artificial rules,it can describe the rich intrinsic information of data better.This paper applies the Extreme Learning Machine based on deep belief network to personal credit risk assessment,and makes an empirical comparison with the traditional Logistic Regression and Gradient Boosting Decision Tree.In view of the imbalance of samples,this paper attempts to use K-BSMOTE to balance the data.Empirical results show that balancing data can effectively improve the classification performance of the model.By comparing and analyzing the above models,it is concluded that the Extreme Learning Machine model based on deep belief network is slightly better than Logistic Regression and Gradient Boosting Decision Tree.