AI · PRISMA-guided systematic review; 240 studies bibliometrically assessed and 61 qualitatively synthesized
Customer Churn Prediction: A Systematic Review of Machine Learning and Deep Learning Methods
Machine Learning and Knowledge Extraction · 2025
Summary
Ensembles dominate; deep learning is growing; class imbalance, interpretability, drift and weak profit-oriented evaluation remain challenges.
For your marketing
Possible founder application
Evaluate churn models with intervention profit, calibration, drift monitoring and explainability—not AUC alone.
Reported result
What the study found
Ensembles dominate; deep learning is growing; class imbalance, interpretability, drift and weak profit-oriented evaluation remain challenges.
Research question
What the authors investigated
What methods and gaps characterize recent ML/DL churn prediction research?
Design
PRISMA-guided systematic review; 240 studies bibliometrically assessed and 61 qualitatively synthesized
Sample and setting
Peer-reviewed churn-prediction studies published 2020–2024.
Scope and transfer
Limitations to keep in view
No formal risk-of-bias/certainty assessment; heterogeneity prevented meta-analysis.
Research sources
Methods/results in the publisher full text