Effective ML Techniques to Predict Customer Churn

Customer churn is one of the most challenging problems that affects revenue and growth strategy of a company. According to a recent Gartner Tech Marketing survey, 91% of C-level respondents rate customer churn as one of their top concerns. However, only 43% have invested in additional resources to support customer expansion. Hence, retaining existing customers is of paramount importance to a company's growth. Many authors in the past have presented different versions of models to predict customer churn using machine learning techniques. The aim of this paper is to study some of the most important machine learning techniques used by researchers in the recent years. The paper also summarizes the prediction techniques, datasets used and performance achieved in these studies for a deeper understanding of the domain. The analysis shows that although hybrid and ensemble methods have been widely successful in improving model performance, there is a need for well-defined guidelines on appropriate model evaluation measures. While most approaches used are quantitative in nature, there is lack of research that focuses on information-rich content in customer company interaction instances, like emails, phone calls or customer support chat records. The information presented in the paper will not only help to increase awareness in industry about emerging trends in machine learning algorithms used in churn prediction, but also help new or existing researchers position their research activity appropriately.