Applicability of Machine Learning Algorithms for Customer Churn Prediction

Abstract

In the digital economy, customer retention is vital for profitability in e-commerce, telecom, and banking. This paper reviews machine learning (ML) and deep learning (DL) architectures for churn prediction (2020–2025). While traditional models offer interpretability, they struggle with high-dimensional data handled by advanced algorithms like XGBoost, ANN, and LSTM. The study emphasizes critical preprocessing, specifically addressing class imbalance through SMOTE and SMOTE-ENN. Findings reveal that hybrid frameworks—such as K-Means combined with SVM— consistently outperform standalone models. However, challenges persist regarding the "black box" nature of models and concept drift. The paper concludes by advocating for Explainable AI (XAI) to transition churn prediction from a reactive technical process to a transparent, profit-aligned business strategy.