Prediction Of Auto Insurance Claim Frequency Based On Machine Learning Algorithm

With the rapid development of automobile industry,the proportion of automobile insurance in the whole insurance industry is gradually increasing.Automobile insurance belongs to the category of non-life insurance,and the pure premium of non-life insurance is the product of the frequency of claim and the average amount paid.Therefore,the modeling of auto insurance claim frequency is particularly important for the pricing of auto insurance products.In this thesis,based on regression tree and neural network algorithm in machine learning algorithm,the claim frequency of an auto insurance claim data set is predicted.First of all,we conducted a descriptive analysis of the data,which was helpful to discover the characteristics of the data.Then,we preprocessed the data,converted the classified feature vectors into numerical values by One-Hot coding,and normalized the continuous feature vectors by Min-Max.Secondly,based on the neural network,the traditional actuarial model is embedded into the neural network by using the embedding layer for the classification feature vectors.Finally,considering the interaction between each feature vector,ICANN model is proposed based on CANN model,and the prediction results are compared with those of traditional actuarial models GLM and GAM.The results show that the machine learning algorithm performs better than the traditional actuarial model on this data set,and the embedding layer may improve the out-of-sample performance of the network.ICANN can intuitively identify the relationship between different input levels,so it is also more stable.