Risk Prediction Of Illegal Fund Raising Of Enterprises Based On Machine Learning Methods

In recent years,illegal fund-raising,relying on Internet platforms,has developed rapidly in the private capital market,bringing risks that cannot be ignored to the safety and supervision of private capital,and social damage is becoming more and more serious.The number and amount of illegal fund-raising cases in China has been increasing year by year,and the situation of cracking down on illegal fund-raising remains severe.How to establish a predictive model based on enterprise information data to determine whether the enterprise has the risk of illegal fund-raising,develop an enterprise holographic profiling system based on the integration of government big data and social big data,and provide accurate profiling and services for the enterprise,which is important for grasping comprehensive and reliable information and timely prevention The risk of illegal fund-raising by enterprises has important significance and value.After studying the relevant literature on the risk of illegal fund-raising of enterprises,using the real data that has been desensitized by CCF BDCI,the company's various information data is analyzed and mined through four machine learning models to predict the risk of the enterprise as an illegal fund-raising enterprise.The classification and modeling of illegal fund-raising enterprises consists of four steps:The first step is data processing.Including the original data to fill in missing values,the quantification of text variables,etc.,to pave the way for subsequent feature selection.The second step is feature selection and unbalanced sample processing.For all the variables obtained after processing,variable correlation coefficients and logistic regression lasso coefficient shrinkage are used to achieve variable screening,and positive samples are resampled to synthesize data through the SMOTE algorithm,which reduces the negative impact of unbalanced samples.The third step is model training and prediction.Four models of Random Forest,XGBoost,Light GBM,and Cat Boost are used in this paper.After parameter optimization of each model,the prediction effects of the four models are comprehensively compared through k-fold cross-validation,and the SHAP value obtained in the model is used as a measurement variable.The basis of importance.In the fourth step,after horizontally comparing the prediction effects of the four models,the four models are weighted by the highest F1 value to obtain the optimal combination model,and the SHAP value is used to provide a strategic reference for preventing the risk of illegal fund-raising by enterprises.Finally,combined with the existing research results,it reviews the shortcomings in the current research,and puts forward relevant suggestions on the establishment of a national unified enterprise illegal fundraising big data risk monitoring system.