Research On The Prediction Of Insurance Payment Based On Logistic Regression Model

In the age of Big Data,more and more insurance companies began to use data mining technology to achieve accurate marketing,and fully tap the value of customers.The renewal of insurance policy payment is one of them.Before the deadline for payment,the premium department of insurance company that is in charge of the renewal of policy of insurance,need to call or visit customers in person.The cost of this approach is too large and it is inefficient.If the insurance premiums can get notification early for warning of difficult charging,they can do a good job in advance to urge the payment of the preparatory work,and improve work efficiency.In order to solve the problem of renewal payment of insurance policy,data is extracted randomly from business system according to CRISP-DM standard at first.Then it is explored and pre-processed to select the appropriate data set.On this basis,logistic regression algorithm is used to construct the renewal model of insurance policy,and the model is evaluated and applied.It is found that the model does not consider the influence of data morphological changes and time factors on the sample under the in-depth analysis,and the model has no capacity for automatic tuning.That is,the model has multiple multicollinearity problems.In the insurance business system,the reason of the multicollinearity problem is the existence of the correlation of data,variables and parameters.According to the theory and model of factor analysis,this thesis proposes a method to deal with multicollinearity problem from the perspective of factor analysis.Before modeling,KMO was used to judge whether the data variables needed factor analysis,then 10 common factors were extracted from the 36 data variables of insurance policy.The data set was transformed,and then the logistic regression model was used to predict payment base on the common factor dataset.The experimental results show that the proposed method of factor analysis can effectively reduce the impact of multicollinearity on the model,compared with the one of principal component analysis.The predictive accuracy of the model constructed with proposed method of factor analysis is higher than that of the model constructed with principal component analysis,can reach more than 86%.