Ensemble classification using balanced data to predict customer churn: a case study on the telecom industry

Today, in addition to reactive methods, companies try to use proactive techniques for the early detection of customer churn. Generally, gaining a new customer is more costly than retaining existing customers. A literature review shows that machine learning is the most common approach for predicting customer churn. This paper examines the factors affecting customer churn in the telecom industry. In this regard, we use data mining classification methods, such as neural networks, K-nearest neighbor, support vector machine, logistic regression, decision tree, and random forest. The results are analyzed using various criteria such as accuracy, precision, recall, F1-score, and ROC curve. This paper examines the mutual effect of data balancing, acceleration algorithms, and high-accuracy classifiers such as neural networks. The most important contribution of this paper is to present a speed-accuracy trade-off in hybrid classifiers for solving real-world problems. It examines the performance of the classifiers before and after data balancing. After determining the successful classifiers, we combine them with the AdaBoost and XGBoost methods. The most effective combinations are identified according to all evaluation criteria. Our results show that the use of a hybrid classifier with the help of AdaBoost and XGBoost significantly improves performance. The combined classification presented in this research is more accurate than the state-of-the-art methods.

Automated summary

This paper examines the factors affecting customer churn in the telecom industry and analyzes them using various data mining classification methods. By evaluating different criteria, this paper demonstrates the trade-off between speed and accuracy in hybrid classifiers and proposes a more accurate combined classification method.