An online Churn prediction model based on CRM with support of heterogeneous data for life insurance
Machine Learning, Classification, Big Data, Online Churn Prediction, Life Insurance.
The customer churn is the loss of a client, that is, when the customer decides to cut ties with the company during a time period. There exist many reasons to churn. For instance, poor service and relationship, acquisition of a wrong product, and tie-in sale. These have a negative impact in company finances that could reach millions of dollars in losses, or even the brand. For instance, the tie-in sale represents about 70% of the client's complaints. The aforementioned problems might lead to loss of millions of dollars. Thereby, diverse industries have been trying to better understand their client's behavior and needs. This is one of the most important key concepts to deal with the customer churn. The telecommunications industry has been investing in new technologies and ways to retain customers since customer retention is approximately 7 times cheaper than win a new one. On the other hand, the insurance industry has not been well explored yet. It has been focused on only one product, auto insurance. However, each insurance product has different behaviors. Thus, a churn prediction model for auto might not work for other products such as life and home insurance. Unfortunately, the life insurance lacks of research in customer churn prediction. It should gain more attention since it is the most widely applicable.
Therefore, this work aims to build an online churn prediction method for life insurance. The method is based on Core Vector Machine (CVM) and Minimum Enclosing Ball (MEB) which must separate geometrically churners from non-churners. The CVM, in essence, is not an online method. However, we intent to use it as basis for our method since it can deal with large datasets and its time efficient when an approximation is applied. Transform and use it as part of our online method is one of the challengers since the proposed solution must run on a Big Data Environment. In order to increase the accuracy of the churn prediction it also must be supported by heterogeneous data on its decision. As the final result, we expect to reduce the (monetary) losses by giving the support in decision making through the identification of the possible churners.