Research On Markov Insurance Pricing Based On Diabetes Risk Cluster

At present,the field of health insurance still uses traditional actuarial models and empirical rates to set the underwriting price.This simple pricing method cannot fully reveal the type of risk of the insured,and often leads to serious adverse selection.With the use of machine learning methods and medical data sets,risk prediction based on data analysis is adopted to distinguish the level of risk of policyholders,and the correct classification of the risk level of policyholders will become an effective way to avoid adverse selection.At the same time,the prevalence of diabetes in my country continues to increase,the onset cycle is long,the complications are serious and the age tends to be younger,which has caused huge economic pressure on the society.Therefore,this research combines machine learning and insurance actuarial methods to develop diabetes insurance that can compensate for the medical expenses incurred by suffering from diabetes.Product diversification can effectively avoid adverse selection problems in practice.The main tasks of this research are:In view of the existence of multi-dimensional discrete and disordered features in medical data sets,the C-Relief algorithm considering sample correlation is proposed,which improves the distance measurement method of discrete disordered variables,and calculates the proportion of discrete feature values in various sample labels.To measure the contribution of features to the degree of sample separation,and then determine the weight of each feature.Focusing on the medical examination indicators of the insured,the C-Relief algorithm is used to eliminate irrelevant features in the medical data set,and the data dimension is reduced.And the obtained feature subset has higher accuracy and better performance.Combining the clustering algorithm and the insurance actuarial model,the differentiated pricing of the group is realized.Through feature selection and feature extraction of low-dimensional features related to diabetes,policyholders are classified into different risk levels based on prior attributes,and the multi-risk state obtained by clustering is combined with the initial multi-state Markov model to achieve diversified pricing,Achieves the differentiated pricing target of the policyholder group of diabetes insurance,makes the insurance premium and the risk level match,and verifies the effectiveness of this method on the medical data set.The clustering considers the influence of feature weight and ambiguity,and realizes the personalized pricing of a single policyholder.Considering the influence of the weight of each feature obtained by feature selection,compared with the previous clustering methods,all feature weights are equal.In this paper,the spatial axis is scaled in the distance measurement to consider the weights of different features to improve the clustering performance.Performance,a weighted fuzzy clustering(WFCM)method is constructed,which takes into account the uncertainty of the insured’s attribution of different risk clusters in practice,expressed by a fuzzy membership matrix,and personalized pricing is calculated by weighting its membership.Experiments show that personalized pricing that considers ambiguity can better reveal the level of risk.