Insurance Risk Prediction Based On Neural Network Model And SVM

Deep neural networks and support vector machines are classic algorithms that are often applied to classification problems in machine learning.Both methods have their own advantages and limitations in the model.The deep neural network calculates the output error by learning the data characteristics,and then realizes the parameter update through the back propagation of the error.After continuous iteration,the training is completed and the classification prediction function is realized.The support vector machine divides the samples by constructing a decision function.In a sense,this decision function is a separation hyperplane,which divides the samples according to the goal of "maximizing the interval".When encountering a scene that is not linearly separable,SVM uses a kernel function to map the vector in a high-dimensional manner,and achieve linear separability in a new space.Compared with neural networks,SVM is more suitable for small sample,nonlinear scenes,is not prone to overfitting,and has certain fault tolerance.But when the sample size is large or the kernel function is introduced,a large amount of calculation memory will be consumed.Therefore,this paper uses two classification methods at the same time to solve the problem of insurance risk classification.This article first introduces the working principle of DNN and the derivation of the algorithm.Then introduced the principle and classification of SVM.Finally,based on the policyholder data of an insurance company in the UK,after data preprocessing,two classification models,DNN and SVM,are used to predict the policyholder's risk level.According to the prediction results of the model,different evaluation indicators are calculated,and the internal results of the two model classifications and the comparison between the models are evaluated.It is found that the SVM model using the RBF kernel function is significantly better than the deep neural network in the classification effect.