Research On Personal Credit Risk Assessment Based On Deep Learning

Personal credit has always been the most important factor for banks to measure individual compliance risk.In recent years,with the increasing demand for borrowing in China,default risk rises which is not only the main risk faced by commercial banks,but also the important factor that contributes to the instability of the entire financial system.How to conduct a comprehensive and accurate personal credit risk assessment is the core link of risk control and the demand of improving the risk management level of commercial banks.Traditional credit risk assessment relies too much on personal credit reporting,which has many shortcomings in the timeliness,comprehensiveness and diversity of data and can't meet the needs of the rapid development of current personal credit of banks.The arrival of the big data provides multisource personal data,enriching personal credit portraits.How to fully integrate and utilize bank big data to conduct a more comprehensive personal credit risk assessment is one of the problems faced by commercial banks when commercial banks gradually accumulate big data resources.While using bank big data,the high-dimensional and sparse characteristics of big data bring difficulties in feature selection,which makes traditional credit risk assessment models not well suited for big data environments.In addition,high noise is common in big data,how to effectively solve personal credit risk assessment under noisy big data is an urgent problem to be solved.At the same time,the problem of unbalanced credit risk assessment sample still exists in the big data environment,which will directly affect the effect of the evaluation model.In order to make better use of the big data of bank for personal credit risk assessment,and solve the problems caused by high dimension,sparse and multi-noise in big data environment,effectively avoid the impact of unbalanced samples on risk assessment at the same time,the goal is to improve the bank's personal credit risk assessment level as a whole.This paper has carried out research work in three aspects combining constructing a collection of personal credit risk characteristics based on bank's big data,credit imbalance sample learning based on generative adversarial network,constructing a personal credit risk assessment method with deep learning based on the cutting-edge deep learning technology in the field of machine learning research and statistical analysis with bank's big data.Specifically,the research work in this paper includes the following three aspects:1.Building a personal credit risk assessment feature set based on bank's big data.The characteristics of general banks used in credit evaluation are relatively simple.The incomplete assessment of personal credit leads to the failure of credit resources to achieve optimal supply and the risk of default.Constructing a personal credit portrait based on bank's big data and using the user's portrait principle make up for the lack of information on traditional assessment characteristics,and alleviate the information asymmetry of personal credit assessment.An example is given to illustrate the complete process of constructing personal credit risk assessment features using big data.Empirical comparison is carried out to verify the relevance of big data characteristics and credit risk,and demonstrate the contribution of bank big data to credit risk assessment results combining statistical analysis and modeling.Differential portraits of different credit groups are obtained through big data analysis,which provides a reference and basis for credit risk assessment using big data.2.The learning of credit imbalance samples based on generative adversarial networks(GAN).There is often a sample imbalance in the credit risk assessment problem.At present,the main methods for solving unbalanced sample classes are mainly random sampling from a few samples.The result of this method is sampling inaccuracy and distortion,which in turn affects the evaluation of the final model.The generative adversarial network is a generative learning method proposed in 2014,mainly used to generate data.This method is first applied to the problem of credit risk imbalance samples,and the Focal-Loss GAN method is proposed based on the original method.Experimental demonstrates that the method of this paper has a better effect in solving the problem of unbalanced samples combining public data sets with bank imbalance credit data.3.The personal credit risk assessment method based on deep learning is proposed which is established on the big data of bank.The difficulties in feature selection of traditional credit evaluation models has occurred in high-dimensional,sparse big data environments.In addition,the high noise in big data can also affect the result of the model prediction.An algorithm called stacked denoising auto-encoder neural networks(SDANN)is proposed to apply to personal credit risk assessment in the bank big data environment regarding the issue above.The deep learning method of this paper is more effective in extracting the essential features of expressing credit objects,and it is better than the traditional feature selection method in big data environment through experimental and visual analysis.In addition,the noise reduction method further enhances the robustness of the model and has a better risk assessment effect.This paper uses big data of bank to construct the characteristics of personal credit risk assessment,which can make a more comprehensive evaluation of the bank's personal credit risk,enrich the research and application of big data in the field of personal credit risk assessment of financial institutions,and provide useful process and method for constructing personal credit characteristics for reference.The unbalanced sample learning method based on the generated adversarial network provides a new research idea to solve the sample imbalance problem,further enriching the application scenario of the generated adversarial network in financial problem.A new solution for personal credit risk assessment in big data environment is given based on the deep learning.It has certain reference and guiding significance about the deep learning applied in the field of finance,especially in the field of credit risk assessment under big data environment.