Research On Risk Prediction Of Illegal Fundraising Based On Machine Learning

With the development of Internet finance,private capital has gained more investment channels,and the market can more easily obtain a large amount of financing.However,the quality of each company’s qualifications is uneven and inadequate regulatory system has provided a good breeding environment for illegal fundraising.Identifying illegal fund-raising behaviors through financial data analysis is currently a challenging issue to maintain financial order,protect investors’ rights and interests and the healthy development of the financial industry.In the research field of illegal fundraising,most scholars currently define illegal fundraising at the legal level and how to prevent it.In terms of artificial intelligence,although financial security is also a relatively hot research field,most of them are aimed at the hotter P2 P fields,money laundering In the field of illegal fund-raising,there are few direct risk predictions for the field of illegal fundraising.At present,most of the risk predictions for illegal fund-raising companies are based on the government data and social data of the enterprise.However,for some illegal fundraising companies that are good at disguising,the criminals will deliberately maintain the company’s operating conditions and financial conditions to attract victims.Therefore,there are still certain deficiencies in the risk prediction of illegal fund-raising with the government affairs data and social data of enterprises.In view of the above situation,this paper proposes a hybrid machine learning algorithm to predict the risk of illegal fundraising for enterprises,not only based on the company’s own risk data,but also adding the company’s corresponding bank transaction records.In this paper,the data of 355 companies collected by Qichacha obtained the relevant characteristics of the company’s qualifications and strength after preprocessing and other related preparations,and obtained the company’s risk assessment characteristics.Combined with the capital transaction flow of each company,the valuable information is extracted through data preprocessing,and the transaction status of each user of each company is extracted,and the CNN neural network is used to train the user’s transaction behavior feature prediction model.Use this model to judge the user behavior of each company,get the number of users who meet the behavior characteristics of illegal fund-raising users in each company,and use this feature to combine with the company’s risk assessment characteristics.In the analysis of modeling experiments,mixed model and single model are used for analysis,and the optimal algorithm is selected,and then the algorithm is improved by combining models.The experimental results show that the addition of the feature of the percentage of users who meet the illegal fundraising behavior of all users can effectively improve the performance of the model.Among them,the combined model of CNN-Light GBM has improved greatly,with an accuracy rate of 91.15% on355 test sample data.,which can be well used in practice..