ADVANCEMENTS IN MACHINE LEARNING FOR CUSTOMER RETENTION: A SYSTEMATIC LITERATURE REVIEW OF PREDICTIVE MODELS AND CHURN ANALYSIS

Abstract

Customer retention has emerged as a critical strategic objective for organizations seeking to sustain profitability and competitive advantage, particularly in highly saturated and dynamic markets. Predictive modeling, driven by machine learning (ML) techniques, plays an increasingly essential role in enabling firms to identify customers at high risk of churn and to implement proactive retention interventions. This systematic literature review provides a comprehensive synthesis of contemporary advancements in ML-based customer retention analytics, focusing on predictive models and churn analysis across diverse industries, including telecommunications, banking, e-commerce, and subscription-based services. Utilizing the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) framework, 112 peer-reviewed studies published between 2015 and 2025 were rigorously selected to ensure methodological rigor and relevance. The review systematically categorizes ML techniques into supervised, unsupervised, and hybrid approaches, with a particular emphasis on widely adopted algorithms such as logistic regression, decision trees, support vector machines, random forests, gradient boosting machines, and deep neural networks, including convolutional and recurrent architectures. In addition, it critically evaluates feature engineering methods, data preprocessing practices, dataset properties, and model evaluation metrics, including accuracy, precision, recall, F1-score, AUC-ROC, and cost-sensitive measures. Special attention is given to emerging research domains, including explainable artificial intelligence (XAI), real-time predictive analytics, transfer learning, and federated learning, which enhance model transparency, adaptability, and privacy compliance. Findings from the review reveal that ensemble methods and deep learning models consistently outperform traditional classifiers in detecting intricate churn patterns, particularly when behavioral, transactional, and sentiment-based features are integrated. The review also underscores the growing importance of interpretable models, post-hoc explanation techniques such as SHAP and LIME, and privacy-preserving methodologies to address challenges related to algorithmic opacity, ethical compliance, and deployment scalability. By mapping the evolution of ML-driven churn analytics, identifying persistent research gaps, and proposing actionable directions for future inquiry, this study contributes to both the academic literature and practical applications in customer retention strategy development.