Research On Auto Insurance Claims Prediction Method Based On Internet Of Vehicle

With the rise of the wave of Internet of Vehicles,the analysis based on customer auto insurance claim data and vehicle driving data becomes possible,which makes my country’s auto insurance industry glow with new vitality under the blessing of new technologies.In order to formulate premiums in a differentiated way,auto insurance companies make accurate predictions on each customer’s accident and claim settlement,and improve the profit margin of the auto insurance company.They mainly conduct research from the two aspects of automobile accident prediction and auto insurance claim amount prediction.With data analysis The method is updated and iterated,and the research method has gradually changed from the traditional generalized linear model to machine learning and deep learning.These studies have the following deficiencies:(1)Usually,the static customer policy data used in the prediction research of auto insurance accidents cannot reflect whether there is a necessary relationship between the auto accident situation and the dynamic driving process of the vehicle.The limitations lead to low prediction accuracy.(2)In the research on the prediction of the amount of auto insurance claims,the existing feature extraction methods often ignore the correlation of features,resulting in low prediction accuracy of the amount of auto insurance claims.In view of the above-mentioned problems in the current research,based on the correlation between the static customer policy data and the vehicle driving data,this paper carried out the research on the automobile accident prediction on the data of the "China Communications Vehicle Networking Vehicle Management System".Carried out research on the prediction of the amount of auto insurance claims,the main research work is as follows.Firstly,a prediction method based on the improved D-S evidence theory for auto insurance accidents is proposed,using the on-board OBD(On Board Diagnostics)terminal in the Internet of Vehicles to collect dynamic data of vehicle driving,and constructing it with static data such as violation information and accident information.Create a new data set;use the advantages of machine learning methods that have low dependence on data distribution to establish a car accident prediction model;use the improved D-S(Dempster-Shafer)evidence theory to integrate multiple machine learning methods to achieve complementary advantages of each method,and Accurately predict whether auto insurance will fail or not.The method was verified on the new data set constructed.Experiments show that the accuracy rate of the automobile accident prediction method based on the improved D-S evidence theory proposed in this paper is as high as 88%,and the precision rate,recall rate,and F1 indicators are also excellent.Secondly,a method for predicting the amount of auto insurance claims based on graph deep learning is proposed.In order to extract the spatial correlation of data from massive dynamic vehicle driving data and static auto insurance claims data,this paper first uses a convolutional neural network.The spatial structure information of adjacent features is extracted.In order to avoid missing the key information of the original data during feature extraction,the extracted features are fused with the original features.Then the fused features are transformed into a graph structure by using the heuristic composition method to realize the correlation expression between features.Finally,the graph convolutional neural network is used to accurately predict the amount of auto insurance claims.The method was verified on the data set provided by a foreign company.The experimental results show that the 1DCNN-GCN model based on graph deep learning proposed in this paper has excellent performance.Compared with random forest and XGBoost,the performance of root mean square error is improved by 3.20.% and 2.22%,and the average absolute error performance is improved by 4.76% and 2.08%.In addition,the generalization of the model was further verified on the data of a domestic company.