Research On The Forecast Of Accumulated Claim Amount Of Auto Insurance Based On Boosting Algorithm

In the course of three times commercial auto insurance rate reforms,insurance companies have gradually increased their voice in auto insurance pricing,and reasonable estimation of auto insurance premiums is the core of enhancing competitiveness and the steady development of the market.The key to determination of auto insurance premiums is to calculate pure premiums.Accurate prediction of the auto insurance accumulated claims amount is the core in the accurate estimation of pure premiums.At present,the Generalized Linear Model is mainly used to predict the accumulated claim amount,but its limitation is that it needs to make assumptions about the data distribution in advance,and the amount and complexity of future auto insurance claims data makes it extremely difficult to assume the distribution of data.Therefore,the accuracy and versatility of the Generalized Linear Model for predicting cumulated claims amount will be unsatisfactory.The Boosting of machine learning algorithm has many good features,especially it does not need to make any distribution assumptions on the data,and it has a good fitting effect for large data sets.Therefore,this article is based on the Boosting to predict the cumulated claim amount of auto insurance and compare it with the Generalized Linear Models.The main research contents are as follows:Firstly,this article summarizes the related application research of the Generalized Linear Model and Boosting to predict the cumulated claim amount by domestic and foreign scholars,and points out the limitations of the Generalized Linear Model and the inevitability of Boosting.Then,construction of the Generalized Linear Model and Boosting is explained,and the XGBoost and GBDT are selected to predict the cumulated claim amount.Subsequently,XGBoost,GBDT,and Generalized Linear Models based on different distributions are used to predict the cumulative claim amount on the selected data set,and the prediction results of each model are compared based on the selected evaluation indicators.The results show that the prediction models of XGBoost and GBDT's cumulated claim amount are better than the Generalized Linear Model in terms of prediction accuracy and other dimensions,and its predictive effect on the cumulated claim amount has been significantly improved.Boosting is more suitable for complex data set than the Generalized Linear Model.Although Boosting's interpretability is not as good as the Generalized Linear Model,and the data reprocessing and parameter adjustment before modeling increases the difficulty of use.But for insurance companies,it is economically beneficial to give up part of the interpretability to improve the versatility of the model and the prediction rate.Finally,prospects and suggestions are made for the promotion of auto insurance accumulated claims amount based on Boosting.