A Targeted Hybrid Model to Customer Churn Prediction in the Insurance Industry (A Case Study)

This study first applies a new customer value model (LRFMPG) in an insurance company that examines the impact of length of the relationship, recency, purchase frequency, monetary value, profit and groups of purchased insurance policies on the valuation of customers. This study employed the analytic hierarchy process to determine the weight of each LRFMPG variable. After the identification and selection of valued customers, development of the churn prediction model is performed with extraction of effective variables and factors on the policyholders’ churn, and then the importance of these factors on the churn of valuable customer is investigated. For this purpose, predictive models are built using different methods (neural networks, decision trees, logistic regression and support vector machine) and the accuracy of the built models has been evaluated, and eventually, a hybrid model has been suggested. The results show that the proposed hybrid model has a higher accuracy than single models in churn prediction.