Research On Risk Prediction Of P2P Online Lending System Based On Deep Learning

With China's economic and financial reform,the traditional lending methods have been unable to meet the loan needs of a large number of small and medium-sized micro enterprises and individuals.The emergence of network lending platform has solved the rapid financing needs of small and medium-sized micro enterprises and individuals.Compared with the traditional domestic lending projects,P2P network loans have some advantages that can not be ignored,such as low requirements for borrowers,fast speed and so on.But at the same time,the network loan relies on the Internet,and there are many problems such as large amount of data and insufficient data authenticity.Therefore,the P2P network loan also has great risks.The traditional rating method can not get accurate risk analysis results.This thesis proposes a model based on stacked noise reduction self-encoder to process the P2P network loan data and effectively improve the accuracy of risk prediction degree.Traditional Stacked Denoising Autoencoders has obvious advantages in risk prediction of P2P system,but the imbalance of data and the poor integrity of user information in P2P system will lead to the problems of low precision,poor robustness and easy to fall into local optimum in deep network training.This thesis proposes a risk prediction method for P2P network credit system using convolution Stacked Denoising Autoencoders as deep network.The deep learning network is used to generate characteristic data matrix,and then machine learning model is used to classify the data.Experiments show that this method can effectively improve the accuracy of risk prediction and solve the problem of data sparseness.Moreover,the flexibility of the algorithm in this thesis is much higher than that of the traditional rating model.The main work of this thesis is as follows:This thesis uses real data from the P2P platform,but the data has the problem of sample imbalance.To solve this problem,data is preprocessed and SMOTE algorithm is used to solve the problem of class imbalance.In order to solve the problem of sparse user data,convolution technology is added to stacked noise reduction self-encoder,and convolution reconstruction is combined with stacked noise reduction self-encoder to effectively solve the problem of inaccurate prediction results caused by data sparsity.In this thesis,masking noise is added to the hidden layer of deep network and the constraint degree of damage handling is introduced.A risk prediction model of P2P online credit system based on in-depth learning is designed and implemented.The data from P2P network loan platform are divided into training set and test set.The improved Stacked Denoising Autoencoders is used to extract the data features of training set.The DTBSVM classifier is used to classify the users,and the credit rating of each user in the prediction set is obtained.The credit risk of the users in the test set is predicted and the reliability of the model is verified.