Research On Customer Segmentation And Loss Prediction Of Automobile Insurance

With the comprehensive reform of auto insurance,the competition in the auto insurance market is extremely fierce.In terms of industry development,China’s auto insurance premium income is gradually slowing down and the profitability of auto insurance business is decreasing.In terms of corporate customer relationship management,vicious competition such as "price wars" have led to the decreasing trust of auto insurance customers,and the problem of customer churn has become increasingly prominent,while the cost of developing new customers is much higher than the cost of maintaining old ones.The cost of developing new customers is much higher than the cost of maintaining the old ones.The loss of customers will further compress the company’s profit,and the auto insurance industry will encounter a "bottleneck".Take JNR Property & Casualty Insurance Company as an example,the company has many problems such as low profit ratio of auto insurance,rapid growth of non-auto insurance premium income,insufficient attention to customer relationship management,and high customer churn rate.With the development of modern information network technology,the company has accumulated a large amount of business data,and it is especially important and urgent to fully explore the value of customers in the data,strengthen customer relationship management and reduce customer turnover.Based on the literature and theories related to auto insurance customer segmentation and churn prediction research,this paper takes JNR Property & Casualty Insurance Company as the research object,firstly,using the ANP(Analytic Network Process)method of expert group decision making and K-means cluster analysis method,and under the guidance of customer relationship management theory,we construct an auto insurance customer value evaluation index system In this study,we assessed the value of auto insurance customers from both current value and potential value and conducted customer segmentation.Secondly,based on the in-depth analysis of the key influencing factors of auto insurance customer churn,the statistical learning methods such as random forest and BP neural network are used to construct an auto insurance customer churn prediction model grouped by customer segmentation categories and conduct a prediction application study.The content,conclusions and innovations of the study are mainly presented as:(1)A customer segmentation model of auto insurance based on customer value was established.Firstly,we refined and designed the auto insurance customer value evaluation index system containing 13 indicators from four dimensions: income value,risk cost,growth value and customer loyalty,and determined the weights of each indicator using the ANP method of expert group decision making,and evaluated and measured the current value and potential value of each customer of JNR Property & Casualty Insurance Company from 2020 to 2021 by combining the data of customer characteristics.Secondly,based on the measured customer value data,the K-Means clustering method was used to segment customers to obtain three customer groups: key value customers,potential value customers and general value customers.(2)An auto insurance customer churn prediction index system was constructed.Firstly,based on the literature research method,we analyzed the key factors affecting the churn of different customer groups from four aspects,such as customer characteristics,vehicle characteristics,insured characteristics and insurance characteristics,and initially selected 28 indicators;then we used the random forest algorithm to select the characteristics for three types of customers,and constructed three types of auto insurance customer churn prediction indicator systems respectively,so as to lay the theoretical foundation for building customer churn prediction models.(3)An auto insurance customer churn prediction model based on customer segmentation is constructed.Firstly,the auto insurance customer churn prediction models were constructed using Logistic Regression,Random Forest,GBDT,XGBoost,Light GBM and BP neural network for three categories of auto insurance customers,namely key value customers,potential value customers and general value customers,and the experimental effects were compared and analyzed.It was found that the BP neural network model has significant advantages in the auto insurance customer churn prediction problem.Secondly,in order to evaluate the effectiveness of the prediction models,comparative analysis of the prediction applications was conducted on the full data set and the segmented data set using the selected BP neural network models,respectively.It is found that the segmented customer churn prediction models have better improvement effect than both before classification.In summary,this study provides ideas and directions for the development of customer churn prediction and retention measures in the auto insurance industry to a certain extent,and provides references for improving customer relationship management,achieving accurate marketing and reducing customer churn in China’s auto insurance industry.