In view of the current event of life insurance agent embezzle insurance premium and fraudulence, although the number of cases is not high, but the fraud patterns and methods continue to evolve and change. It is difficult to effectively block the agent 's abnormal behavior at the moment, if there is no advance warning. Although it is a single case of embezzlement of premium fraud may still bring financial losses and damage to the company's goodwill. The balance between the two needs to be constantly dynamically corrected, so it is imperative to prevent in advance.
We used quantitative research and data analysis as well as artificial intelligence to establish immediate analysis about the types and models, the conduct detection and monitoring analysis of misappropriation of premiums, to analyze the behavior data of existing policyholders and agents. It will improve the weekly density and operational efficiency of various prevention actions, block the expansion of risks and achieve deterrence effects.
According to the result of the study, the factor of the influence between the three Machine Learning models are Ratio of the Automatic Premium Loan (within a year) and Ratio of the Policy Loan (within a year). These two factors are the key influence of forecasting mechanism that why agents embezzled insurance premium. Besides, there are some factor have significance of the three Machine Learning models, such as court compulsive deductions within five years, seniority, level of the contract, Ratio of the Automatic Premium Loan (past years), Ratio of the Policy Loan (last three years) , Ratio of the Automatic Premium Loan (last three years) and Ratio of the agent’s Automatic Premium Loan (past years). As a result of the three Machine Learning models classification, Support Vector Machine is the optimal one, next is Random Forest but Decision Tree performs less prominent of the three ways.