Research And Application Of Online Lending Platform Credit Evaluation Model Based On Deep Learing

With the rapid growth of Chinese economic industry,the development of the financial industry is becoming diversified,online loan platform has become one of the best development momentum for filling the market gaps of microcredit user.With the implementation of new regulations on the supervision of online loan industry in 2017,the credit risks of online loan platforms have become the most important risk factor affecting its development.Therefore,in order to make the online loan industry step into a healthy and stable development path,it is necessary to improve the ability of online loan platforms to control credit risks.Currently,most domestic online loan platforms mainly adopt the model systems based on traditional machine learning or rule engine to deal with credit evaluation,but the final forecasting result is not ideal,the former evaluation way is very subjective,while the latter,due to consider the training efficiency,use small sample set,leads to too high model fitting and lacks wide applicability.With the extensive application of deep learning,it provides a new direction for the construction of online loan platform credit evaluation.Therefore,this thesis proposes a credit evaluation model of online loan platform based on deep learning technology and constructs corresponding index system,hoping to provide new ideas for the credit evaluation research of online loan platform.This thesis mainly designs and implements the credit evaluation model of online loan platform based on deep learning technology,which is mainly composed of three parts : improved DBN unsupervised feature learning,Softmax classification and Using artificial bee colony algorithm instead of BP algorithm for global parameter optimization.The model first uses DBN for pre-training for feature learning,and initially calculates the initial parameters of the model.In DBN pre-training,this paper introduces the momentum learning rate to improve the anti-data oscillation ability in the parameter optimization process,and then after the DBN model.A layer of Softmax classifier is added to process the prediction of multiple credit classifications.Finally,the artificial bee colony algorithm is used to fine-tune the global parameters of the whole model.Compared to the traditional neural network and model before improvement,this model overcomes deep neural network's falling into local optimum situation easily for its random initialization parameters and data oscillation,and enhances the training efficiency,improves the prediction accuracy.Compared with the dichotomous problems of traditional credit evaluation process,this model further provides the prediction ability to more credit classification,making business meet multiple positioning for a variety of credit people.The artificial bee colony algorithm replaces the original BP algorithm is a new attempt.The global optimization effect and convergence speed of the artificial bee colony algorithm is better than the BP algorithm.The experimental process of this thesis mainly used the real business data of an online loan platform,summarized the selection principles of credit evaluation index by referring to relevant literature,selected 38 indicators with representative characteristic and quantified the qualitative indicators,and then retained 35 characteristic indicators after screening by T test.Based on the completion of index system building of the credit evaluation model of online loan platform,23890 groups of sample data were selected from the business data,samples were divided into 19112 groups of training sample,4778 groups of test sample,the model was trained and tested.Compared to the predicting result of the existing credit evaluation system based on rule engine and traditional machine learning and DBN-Softmax model before improvement,it is verified that the model improved the prediction accuracy of credit evaluation of online loan platform,it is proved that the model is feasible and practicable,it provides practical reference for online loan platform to apply deep learning technology.