A problem-based review in churn prediction model

Abstract

Customer churn prediction is crucial to retain customers and manage business for decision maker. This paper focuses on reviewing existing problems and challenges in investigating telecommunication customer churn, which will help researchers understand the essential factors that affecting the performance of a prediction model. Based on the researches published from 2017 to 2021 in the domain of customer churn prediction, two research questions were proposed, which are (i) what are the issues and challenges faced by researchers in the territory of churn prediction model (CPM); (ii) what are the problems in customer churn prediction model addressed by most relevant studies. In terms of a problem-based literature review, this work indicates that data sparsity, feature selection with bias, imbalanced class distribution and evaluation metrics are major challenges faced by most researchers in CPM. Also, it shows the problems of the inefficient models are due to complex computation, high computation time and yields to unsatisfactory prediction results. In this paper, CPM is discussed focusing on problems and challenges faced by researchers exclusively in telecommunication industry, which is noted as the limitations of this work. For future research, a systematical review regarding data mining and methodology in different business sectors will be explored.