Research On The UBI Premium Based On XGBoost Algorithm

With the development of economy,China's motor vehicle is rising,and correspondingly,the motor vehicle insurance market share is increasing.The auto insurance premium is one of the core issues of auto insurance business,mainly based on the generalized linear model.The generalized linear model is effective to some extent,but it needs to make assumptions about data distribution in advance.In practice,it is very difficult to fit data to a specific distribution.In addition,traditional premiums usually only consider people,vehicles,environment and other static factors,with the development of data science,many factors that not included in the system previously,such as many bad driving behaviors and driving habits,should be built into the premium system,and this will allow the data distribution more complex.It can be predicted that the applicability of the generalized linear model will become worsen in the near future.In recent years,the machine learning is more popular.It has many good characteristics,especially does not need to make any assumptions about the distribution of data,and directly obtain the law from data for prediction,which provides us a good development direction for the auto insurance premiums in complex situations.Therefore,this paper introduces the XGBoost algorithm to predict the cumulative claim amount and the risk probability of pure premium based on people and vehicles.Compared with the traditional generalized linear model method,we find that the results obtained by XGBoost algorithm are valid even without considering the prior distribution of data.This shows that XGBoost algorithm has a broad prospects.After the improvement,it is expected to become the preferred method.In the model of UBI(Usage Based Insurance)rate coefficient,we are fully considering all kinds of driving behavior,especially bad driving habits,the driving behavior scoring model is established with above factors,and the entropy weight-analytic hierarchy method based on game theory with Nash equilibrium is used to assign weights to various factors,it overcomes the subjectivity of AHP and improves the applicability of entropy weight method.We found that the above scoring model has a significant impact on premiums,and different policyholders are charged different premiums according to their driving behaviors,making the premiums more reasonable.In turn,reasonable premiums also helps regulate the policyholder's driving habits,reduce accidents,and promote the benign development of the auto insurance industry.