This paper aims to systematically review research progress in Customer Churn Prediction (CCP). It seeks to construct an intelligent literature analysis framework and summarize the technical evolution and core methodological issues within the domain, providing a comprehensive reference for both theoretical and practical applications.
The study designs an intelligent literature analysis framework integrating Retrieval-Augmented Generation (RAG) with local large language models (LLMs). It systematically analyzes and condenses content from 370 Chinese and English academic publications (2003–2025), focusing on core technical issues such as class imbalance and model construction.
The review summarizes the technical evolution from statistical models to machine learning, deep learning, and ensemble methods. It reveals key development trends, including multi-objective optimization, dynamic adaptability, and enhanced interpretability in CCP research.
This paper validates the efficiency of an intelligent-driven review paradigm. It provides a panoramic synthesis of CCP research, offering valuable references for theory and practice. It also identifies future directions, such as interpretable models and dynamic learning mechanisms, contributing both methodologically and substantively to the field.
