Research On Application Of Kernel Extreme Learning Machine In Insurance Number Of Claims

Auto insurance in the community has been widely concerned,and has an important position in the insurance company,so predicting the number of claims for automobile insurance is one of the important research of non-life insurance actuarial theory.Auto insurance data mainly includes the frequency of claims,the number of claims and the claim strength,so the number of claims for auto insurance is the basis for determining the auto insurance rate.However,there are certain conditions in the process of using the current rate determination method,which limits the scope of the use of the model.For the problems existing in the forecasting of the traditional rate determination model,this paper puts forward the forecasting method of the number of claims for the auto insurance based on the Kernel Extreme Learning Machine,establishes the forecasting model based on the Kernel Extreme Learning Machine theory,,and analysis the forecasting ability of various models,provides new ideas for the actuarial person.The main research contents are as follows:First of all,the methods of several traditional for forecasting the number of auto insurance claims are introduced.In the traditional number of auto insurance claims analysis method,the Generalized Linear Model is often used,but the method needs to assume that the explanatory variable is related to the response variable and needs a lot of empirical data,so there are some problems in the practical application.In addition,Artificial Neural Network model has some disadvantages in the use of the process,including large interference by human factors,slow learning,the initial parameter selection sensitive,poor generalization ability,easy to fall into the local optimal solution and other shortcomings.Then,we use the Kernel Extreme Learning Machine theory to establish the forecasting model.In order to solve the problem that the threshold of the input sample and the hidden layer node threshold are random assignment,which leads to the poor computational stability of the Extreme Learning Machine network,it is proposed to use the Kernel Extreme Learning Machine model.For the parameter selection problem of the Kernel Extreme Learning Machine model,a Quantum Particle Swarm Optimization algorithm with stronger global search capability is proposed in order to optimize the network structure parameters,which effectively solves the problem that the Particle Swarm Optimization algorithm has a long computation time in the calculation process and is easy to fall into the local optimal and other issues,so the algorithm improves the stability of the forecast model.Finally,the third party motor insurance in Sweden is used to compare the results of several models which include Kernel Extreme Learning Machine model,Generalized Linear Model and Artificial Neural Network model.By comparing the prediction results obtained by various models,it is concluded that the Kernel Extreme Learning Machine model based on Quantum Particle Swarm Optimization has the best effect on the auto insurance data.