A survey on machine learning methods for churn prediction

The diversity and specificities of today’s businesses have leveraged a wide range prediction techniques. In particular, churn is major economic concern for many companies. purpose this study to draw general guidelines from benchmark supervised machine learning techniques in association with widely used data sampling approaches on publicly available datasets the context prediction. Choosing priori most appropriate method as well suitable classification model not trivial, it strongly depends intrinsic characteristics. paper, we behavior eleven semi-supervised methods seven sixteen diverse churn-like datasets. Our evaluations, reported terms Area Under Curve (AUC) metric, explore influence characteristics performance studied methods. Besides, propose Nemenyi test Correspondence Analysis means comparison visualization between algorithms, Most importantly, our experiments lead practical recommendation pipeline based an ensemble approach. proposal can be successfully applied