Research On Anti-fraud Of Motor Vehicle Insurance Based On Machine Learning Method

Since the 21 st century,with the continuous development of China's economy and the improvement of people's living standards year by year,the number of domestic motor vehicles has also broken through new highs.At the same time,road traffic accidents caused by motor vehicles are increasing.As an important source of income for property insurance companies,auto insurance business occupies more than 50% of the property insurance market.According to statistics,about 20 percent of auto insurance claims have the potential for fraud,but less than 3 percent of fraud is prosecuted,and the form of fraud is increasingly serious.The comprehensive reform of auto insurance in 2020 also proposes the goal of "lowering prices,increasing insurance coverage and improving quality".In this policy context,machine learning technology can be used to improve the anti-fraud level,and the efficient fraud identification ability can reduce the insurer's expenditure and maintain the benign operation of the auto insurance industry.By combining theoretical analysis with empirical analysis,this paper first summarizes the forms and characteristics of auto insurance fraud,expounds the harm of auto insurance fraud to the insurance industry and society,and summarizes the common anti-fraud methods of auto insurance in recent years.Secondly,it analyzes the information asymmetry,incomplete contract,utility theory and other economic theories related to auto insurance fraud.In the game analysis of auto insurance fraud,improving the rate of fraud identification can effectively reduce the fraudulent behavior of the insured.Machine learning,through technology empowerment,can effectively improve the identification efficiency of auto insurance fraud.After summarizing the theory of machine learning,this paper analyzes the necessity and feasibility of machine learning to help auto insurance anti-fraud.With the support of data,technology and policy,machine learning technology can enhance the efficiency of auto insurance anti-fraud system.In the empirical analysis,based on the insurance company b province branch2019-2020,80428 cases of 38 d sample data of the original car insurance case(4% 96%honest case,fraud)empirical modeling,after data cleaning,significance analysis,equilibrium under sampling steps,such as chose eight significant strong characteristic variables.According to the characteristics of the data set in this paper,this paper selects five machine model algorithms including logistic regression with supervised learning,decision tree,support vector machine,random forest and XGBoost,compares the advantages and disadvantages of the algorithms,and briefly introduces the basic principles of each algorithm and corresponding important parameters before starting modeling.In this paper,the training set and test set are allocated in a ratio of 7:3.The results show that the recognition ability of traditional machine learning algorithms such as logistic regression is average,while the recognition ability of random forest and XGBoost algorithms using ensemble learning is strong,and the AUC score of model evaluation reaches 0.94.The empirical chapter adopts the statistical learning method of machine learning,but it still needs to explore the value of machine learning technology in the field of insurance anti-fraud,and learn from the experience of machine learning technology in practice.This paper lists four cases of anti-fraud insurance using machine learning technology at home and abroad,among which "Wind Listener" of China Pacific Insurance Skillfully uses speech recognition technology to identify fraud information;American Lemonade Insurance Company uses computer vision and natural language processing technology to improve anti-fraud efficiency.Through the above cases,we draw lessons for future development.Therefore,it is feasible for machine learning technology to help auto insurance anti-fraud,but we should continue to explore the value of machine learning in speech recognition,natural language processing and other aspects.Improve the dimension and accuracy of data collection,strengthen research on new technologies and algorithms;Promote multi-party data sharing and interoperability;Strengthen supervision of machine learning technology to prevent risks arising from new technologies.Let machine learning technology to fight auto insurance fraud safely and efficiently,and make a contribution to the auto insurance anti-fraud system.This is of positive significance to the healthy development of China's auto insurance industry and the establishment of China's credit system.