Machine Learning Approaches to Customer Churn Prediction in Subscription Markets

Abstract

This article examines how machine learning has been applied to customer churn prediction in subscription markets such as telecommunications, media streaming, Software as a Service (SaaS), and financial services, where recurring revenue makes customer retention critical. It asks which algorithms and data practices are most commonly used, how they handle issues like class imbalance and high-dimensional features, and whether models are evaluated in terms of business value as well as statistical accuracy. The study employs a systematic review of peer reviewed articles published between 2017 and 2021, synthesizing evidence across telecom, broadband, banking, and other subscription contexts. The findings show that tree-based ensembles, stacking, and deep learning generally outperform traditional statistical models, especially when paired with targeted feature engineering and imbalance handling techniques. The article discusses these patterns by comparing sectors, methodological choices, and evaluation criteria, and concludes that future research should integrate economic objectives, temporal dynamics, and interpretability to make churn models more actionable for subscription businesses.