Home / Current Issue / Paper 1712929
Data Analytics for Proactive Customer Retention
Subject area: Science,Engineering and Technology · Area of research: Data Analytics & Machine Learning
Abstract
Customer churn, or loss of a client, is an issue in telecommunication, overseeing an account, and e-commerce. Acquiring new clients is a part more costly than holding existing ones. Thus, churn prediction becomes a commercial technique. Classic strategies such as logistic regression, decision trees, and random forests are a extraordinary beginning but fall short in addressing to course imbalance, moving behaviors, and requiring actionable outcomes. Recent studies have advanced churn prediction in three ways. First, ensemble approaches such as XGBoost, LightGBM, and CatBoost provide palatable execution, particularly when utilized in conjunction with oversampling strategies such as SMOTE and ADASYN. Second, hybrid deep learning models that combine CNNs, BiLSTMs, and attention mechanisms can superior learn adjacent, progressive, and global features with advanced recall and F1 scores. Third, online learning approaches allow models to diligently learn in real-time of progressing client behavior. This article describes these progressions and proposes a system for proactive upkeep utilizing data analytics. It is found that ensemble models have over 85-90% balanced accuracy, and hybrid profound models provide excellent execution. Versatility, feature importance, and interpretability are found to be of crucial importance for being of practical use.
Keywords
Customer churn, Hybrid Approaches, Deep Learning, XGBoost, LightGBM
References
[1] F. M. Alotaibi and A. U. Haq, “Customer churn prediction in telecommunications using ensemble machine learning and deep learning techniques,” Engineering, Technology & Applied Science Research, vol. 14, no. 2, pp. 8606–8612, 2024.
[2] M. R. Maulana and F. Hidayati, “Comparison of Random Forest and XGBoost for credit customer churn prediction,” Indonesian Journal of Data and Science, vol. 2, no. 1, pp. 15–25, 2025.
[3] H. Zhang, Y. Wang, and X. Liu, “Deep learning-based consumer behavior analysis and application research,” Journal of Physics: Conference Series, vol. 2253, p. 012019, 2022, doi: 10.1088/1742-6596/2253/1/012019.
[4] X. Zhang, H. Li, and Q. Zhao, “Customer churn prediction model based on hybrid neural networks (CCP-Net),” Scientific Reports, vol. 14, p. 79603, 2024, doi: 10.1038/s41598-024-79603-9.
[5] Banerjee and R. Singh, “Mitigating class imbalance in churn prediction with ensemble methods and SMOTE,” Expert Systems with Applications, vol. 240, p. 122610, 2025, doi: 10.1016/j.eswa.2025.122610.
[6] M. Khan and M. Yousaf, “Enhancing customer churn analysis using replay based continual learning and stacked ensembles,” Conference Paper/Preprint, 2025.
[7] P. Singh and A. Verma, “Improved decision tree, random forest, and XGBoost algorithms for telecom churn prediction,” International Journal of Computer Applications, vol. 183, no. 45, pp. 1–8, 2024.
How to cite this paper
@article{1712929,
author = {Ranveer Kumar Singh, Shivam Chaudhari, Rajesh Raghav, Saranya Raj},
title = {Data Analytics for Proactive Customer Retention},
journal = {Iconic Research And Engineering Journals},
year = {2025},
volume = {9},
number = {6},
pages = {1376-1381},
issn = {2456-8880},
url = {https://www.irejournals.com/formatedpaper/1712929.pdf},
abstract = {Customer churn, or loss of a client, is an issue in telecommunication, overseeing an account, and e-commerce. Acquiring new clients is a part more costly than holding existing ones. Thus, churn prediction becomes a commercial technique. Classic strategies such as logistic regression, decision trees, and random forests are a extraordinary beginning but fall short in addressing to course imbalance, moving behaviors, and requiring actionable outcomes. Recent studies have advanced churn prediction in three ways. First, ensemble approaches such as XGBoost, LightGBM, and CatBoost provide palatable execution, particularly when utilized in conjunction with oversampling strategies such as SMOTE and ADASYN. Second, hybrid deep learning models that combine CNNs, BiLSTMs, and attention mechanisms can superior learn adjacent, progressive, and global features with advanced recall and F1 scores. Third, online learning approaches allow models to diligently learn in real-time of progressing client behavior. This article describes these progressions and proposes a system for proactive upkeep utilizing data analytics. It is found that ensemble models have over 85-90% balanced accuracy, and hybrid profound models provide excellent execution. Versatility, feature importance, and interpretability are found to be of crucial importance for being of practical use.},
keywords = {Customer churn, Hybrid Approaches, Deep Learning, XGBoost, LightGBM},
month = {December},
doi = {https://doi.org/10.64388/IREV9I6-1712929}
}