Research On Auto Insurance Fraud Recognition Based On Machine Learning Model

In the last few years,Chinese insurance industry has developed rapidly.Auto insurance is a large proportion of property insurance,but according to statistics,about 20% of auto insurance claims are suspected of fraud.Auto insurance fraud is one of the important reasons for the high comprehensive cost rate of property insurance.serious harm to the company.At present,most insurance companies mainly rely on manual review to identify auto insurance fraud,which greatly increases the cost of insurance companies.With the advancement of machine learning technology,it has become possible to apply machine learning methods to the field of auto insurance fraud identification.The use of machine learning technology for auto insurance fraud identification has the advantages of lower cost and higher accuracy than manual work,which will effectively improve insurance companies.Economic benefits,thereby improving the status quo of the high rate of auto insurance fraud.The auto insurance fraud in this article mainly refers to the behavior that violates the law or the provisions of the insurance contract.This paper builds a sample equalization model based on Smote and single-class support vector machine to solve the problem of unbalanced auto insurance claim data.In order to obtain a vehicle insurance fraud identification model with better prediction effect,this paper uses the training set after equalization based on Smote,Smote and single-class SVM sample equalization model,respectively,for logistic regression,decision tree,random forest and Light GBM models.Then,the decision tree,random forest and Light GBM models with better prediction effect were selected,and the three models were fused using the Stacking ensemble learning model,and finally their prediction effects were compared and analyzed.The final results show that after processing the training set through a sample equalization model based on Smote and single-class SVM,the prediction effect of each machine learning model is better.Compared with the prediction effect of a single model,the prediction accuracy of the auto insurance fraud identification model based on the Stacking ensemble learning model is higher,whether for fraudulent claims or non-fraudulent claims.The better prediction effect of this model can better assist insurance company reviewers in reviewing auto insurance claims,which can reduce the review cost of insurance companies,thereby safeguarding the legitimate rights and interests of insurance consumers.