Investigating the churn of life insurance customers using data mining methods(A case Study: One of the Iran’s insurance companies)

Customer retention is always considered as the most important principle in all industries, and the insurance industry is no exception. During the recent years in the Iranian society, with the increase in the sale of life insurance policies, the retention of insurance customers has become more and more important to the managers and experts of the insurance industry so that they can keep a wide range of customers. Nowadys, creating a sense of satisfaction in life insurance customers as a management art has been noticed by insurance companies. The more customers the insurance company can keep happy, the less they worry about redemptions and exits. The main goal of this research is to implement data mining methods in predicting customer churn and identifying factors affecting customer churn in the life insurance products of one of Iran's insurance companies. The purpose of customer loss forecasting is to identify the desired class or class related to insurance policies that are suspended or canceled at the request of the policyholder before the end of the insurance coverage period.

In this paper, we have tried to classify life insurance customers based on abdication or non-withdrawal using data mining algorithms such as random forest, decision tree, logistic regression and neural network. The data used in this research include the information of life insurance policies of an insurance company in 2019 in Tehran province, which has a high and appropriate share in the portfolio of the insurance industry. To evaluate and compare these 4 methods, different criteria will be used. In the field of data mining, and in particular the problem of classification, the confusion matrix as a special tabulation makes it possible to visualize the performance of an algorithm. The confusion matrix shows how many true and false predictions have been made for each class, and based on these values, different criteria for classification evaluation and accuracy measurement can be defined.

The results of the research show that random forest, decision tree, logistic regression and neural network algorithms have high performance in predicting the class related to customer churn. Based on the results of the research, the probability of re-buying was better in women and people with high-risk jobs and older age. On the other hand, people who initially paid the insurance premium annually or chose a lower premium and a higher percentage of capital change factor and capital risk, the probability of their redemption was less.

Considering that life insurance is usually long-term and also considering the liquidity needs of customers and the current economic conditions of the society, insurance companies should pay more attention to life insurance customers. Also, they should put fidelity programs in order to keep customers on their agenda by continuously monitoring the customer's behavior during the insurance policy.