Customer Churn Prediction in Telecommunication Using Machine Learning: A Comparison Study

Telecommunication operators need to accurately predict the customer churn for surviving in the Telecom market. There is a huge volume of customer records such as calls, SMSs and the use of Internet. This data contains rich and valuable information about costumer behavior and his/her pattern consumption. Machine learning is a powerful tool for extraction of costumer information that can be useful for churn prediction. Although several researchers have studies some types of machine learning methods, but, there is not any work which assess different methods from various point of views. The aim of this work is to assess the performance of a wide range of machine learning methods for churn prediction in the form of a comparison study. In this paper, various machine learning methods consisting of 7 classifiers, 7 target detectors, 10 feature reduction methods containing 4 feature extraction algorithms and 6 feature selection ones are discussed. The performance of these methods are experimented on three Telecom datasets with 6 evaluation measures. The results show that the random forest and feed-forward neural network beside the genetic algorithm outperform other competitors. The superior methods achieve 97%, 62% and 93% prediction accuracy in BigML, kaggle and Telco customer churn datasets, respectively.