Customer Churn Analysis

Abstract

In this study, a system has been developed to predict customers who may leave the private pension system. For this purpose, a training data set was formed by combining the churn contracts in the previous months or years with nonchurn contracts for both classes equally. In the train data set, attribute selection was made and learning models were created. The classification system,training, is made consistent with the addition of new data every month. Using the model that was trained with the cumulative training data set, customers who are likely to leave for the next month are estimated. In the classification, the average test accuracy for the 18 months from March 2018 to September 2019 was %99.01. In the separation estimation study, precision and recall are important parameters because of the imbalance between the classes. In this study, the average recall was calculated as %98.99 and the average precision was calculated as %6033