A Research On The Application Of Decision Tree And Generalized Linear Model In Auto Insurance Pricing

In recent years,generalized linear model(GLM)is widely used in private auto insurance classification pricing in auto insurance market in and out abroad.Decision tree is one of the big data methods,and is also an important method for classification.Without any supposition on the structure of the model beforehand,it can directly explore the relationship between the target variable and the related variables.The application of decision tree method in auto insurance is booming.In this paper,by using this two methods and related SAS and R software,the auto insurance policy database of a property insurance company between the claim data and related risk factors is set up and discussed.Firstly,one-way analysis is developed and related variables on frequency,severity and loss ratio are found.Secondly,the GLM models on frequency and severity are set up and improved respectively,the relative importance of the variables is compared and the values of the parameters are calculated.Thirdly,the decision tree models and randomForest models on frequency,severity and loss ratio are developed and pruned respectively.The accuracy of the models is analyzed and the effective rules are obtained.Finally,the advantage and disadvantage of the two models are analyzed.The characteristic of the models,the choosing of the variables and the stability of the models are compared.Although the two methods are different at first glance,we find that decision tree method is almost as efficient as GLM model method in the choosing of the variables,the accuracy of the classification and the stability of the models in the database.Besides,the decision tree method is simpler and easier than GLM.In one word,the decision tree method has good application view in auto insurance company,especially in the field of choosing the target customers,setting up the effective underwriting model,inspecting the high risk portfolio and controlling the claim risk.