P2P Network Loan-based Results Based On Automatic Machine Learning Research And Application Of Predictive Models

In the current era of Internet finance,in the face of important financing issues affecting the development of small and medium-sized enterprises,P2 P network credit came into being.For the P2 P network credit platform,how to better grasp the quality of funds,centralized management of funds,and risk management are the key concerns of the P2 P platform.In this context,this paper explores the problem of P2 P network credit feature loan result prediction,and conducts in-depth research on sample feature extraction,expert sample unbalanced anti-neural network construction process and automatic machine learning model.Firstly,based on the P2 P network credit text data,this paper transforms the text data into numerical features based on regular expressions,and divides the data into personal loan data and corporate loan data through the existence of business hours.Secondly,aiming at the problem of positive and negative case imbalance of the sample,an anti-neural network constructed by a model similar to the generation model of the automatic encoder and the discriminant model of the bp neural network is established,and the oversampling structure of the sample is constructed.Then,the paper explores the P2 P network credit loan loan prediction model based on the automatic machine learning module auto-sklearn and proves the feasibility of automatic machine learning in credit data application.Finally,by using automatic machine learning and experimental analysis of other classification models,for the automatic machine learning model,although the classification accuracy in the enterprise credit data is slightly inferior to the integrated classifier,the recall of personal credit data and corporate credit data.In terms of rate,the performance of the standard automatic machine model shows obvious advantages.Especially from the comprehensive ability of precision and recall rate,the stability of automatic machine learning in this experimental data is the best.