Customer Subscription, Segmentation, and Retention Using Machine Learning Models

Abstract

In the current era of data-driven marketing, effectively managing customer subscription, segmentation, and retention has become essential for sustaining business growth and enhancing user engagement. This study introduces a unified, explainable framework that concurrently addresses all three dimensions using a blend of machine learning and deep learning methodologies, tailored for practical deployment across diverse domains. For customer subscription prediction, we utilize the Bank Telemarketing dataset, comparing the performance of traditional ensemble learning models with a proposed deep learning model that incorporates data balancing techniques. The proposed model deserves a significant improvement, reaching with 93.00% prediction accuracy. In customer segmentation, we employ a customer purchase-related dataset and evaluate two clustering techniques KMeans and Gaussian Mixture Model (GMM) where GMM demonstrates the most effective separation of customer behavior clusters. Feature scaling is applied on the LRFMSQ (Length, Recency, Frequency, Monetary, Satisfaction, Quantity) attributes to ensure uniformity and improve clustering performance. For customer retention prediction, Telecom data was used, comparing the performance of the proposed method with the existing.The proposed method outperforms the existing. This proposed method consists of a Generative Adversarial Network (GAN) for class imbalance and ConvLSTM with attention mechanism, and Grey Wolf Optimization (GWO).

Introduction

The rapid evolution of technology has shifted customer expectations, compelling businesses to adapt by analyzing customer behavior, including retention, segmentation, and subscription. Customer subscription prediction, especially in banking, helps improve market position and loyalty. However, challenges such as targeting the right customers and commitment aversion exist. Traditional machine learning models struggle with complex patterns due to reliance on manual feature engineering.

Customer retention is crucial as retaining customers is less costly than acquiring new ones and increases loyalty and referrals. Advanced methods using Generative Adversarial Networks (GAN), ConvLSTM with attention mechanisms, and Grey Wolf Optimization (GWO) show promise in telecom churn prediction.

Customer segmentation groups customers by behavior to enable targeted marketing. This study uses Gaussian Mixture Models (GMM) with scaled behavioral features (Length, Recency, Frequency, Monetary, Satisfaction) to achieve more accurate segmentations than traditional methods like KMeans.

Recent works across telecom, banking, and finance emphasize machine learning, ensemble models, and hybrid approaches to enhance churn prediction, retention, and segmentation, highlighting algorithms like XGBoost, Random Forest, GAN-LSTM, and ensemble methods.

Methodology:

• Datasets from banking, telecom, and automotive industries were used.

• Preprocessing included normalization, encoding, and handling class imbalance with SMOTE and GAN.

• Models:

  • Customer subscription predicted via a deep neural network (multi-layer perceptron) with dropout, batch normalization, and LeakyReLU.

  • Retention modeled using ConvLSTM with attention and GWO optimization.

  • Segmentation performed using KMeans and GMM on LRFMS behavioral features.

• Training incorporated techniques like early stopping and learning rate reduction.

Results:

• Customer subscription model achieved 94.17% accuracy, with high recall (97.9%) and balanced precision and F1-score.

• Retention model scored 90.28% accuracy with strong precision and recall (~91-93%).

• Segmentation: GMM outperformed KMeans, with nearly 98% accuracy, showing better handling of complex customer clusters.

Conclusion

This study on customer subscription approach proposed a deep learning-based Multilayer Perceptron (MLP) model to predict customer subscription behavior in bank marketing campaigns, achieving a high accuracy of 94.17% along with strong precision, recall, and F1-score values. The model effectively identified key predictors such as call duration, previous campaign outcomes, and past contact status, offering meaningful insights to improve marketing strategies. For future work, integrating explainable AI tools like SHapely Additive exPlanaitons (SHAP) and Local Interpretable Model-Agnostic Explanations (LIME) can enhance transparency and trust, especially among business users. Additionally, incorporating advanced models such as BERT and LSTM could further improve prediction accuracy by capturing context from customer messages or sequential behavior data. These enhancements would be especially useful in real-time decision-support environments like CRM systems, where the model could adapt dynamically to evolving customer patterns. Expanding the model to handle multi-class outputs and time-series trends can also provide more granular and forward-looking insights. Overall, the current work establishes a strong foundation for intelligent marketing systems and opens promising directions for future development in predictive analytics. For telecom retention prediction, ConvLSTM method with GAN, Attention mechanism, and GWO yields 90%. By the improvement of performance of GAN, better results might get achieved which was the limitation of this work. The customer segmentation strategy implemented in this analysis successfully categorized consumers based on behavioral attributes using the LRFMS model. Clustering techniques such as GMM and KMeans were used to uncover actionable insights, allowing businesses to identify high-value, loyal, and at-risk customer groups. Although the dataset was sourced from the automotive spare parts sector, this methodology is highly adaptable and can be applied to customer data across various industries, including retail, e-commerce, and services. These insights can be leveraged to design targeted marketing strategies, improve customer retention, and drive overall business growth. Looking ahead, integrating time-based clustering approaches like TimeSeriesKMeans may enable a better understanding of changing customer behavior and support timely business decisions.

References

[1] Thangeda, R., Kumar, N., & Majhi, R. (2024). A neural network-based predictive decision model for customer retention in the telecommunication sector. Technological Forecasting and Social Change, 202. [2] Singh, P. P., Anik, F. I., Senapati, R., Sinha, A., Sakib, N., & Hossain, E. (2024). Investigating customer churn in banking: A machine learning approach and visualization app for data science and management. Data Science and Management, 7, 7–16. [3] Boozary, P., Sheykhan, S., GhorbanTanhaei, H., & Magazzino, C. (2025). Enhancing customer retention with machine learning: A comparative analysis of ensemble models for accurate churn prediction. International Journal of Information Management Data Insights, 5. [4] Arshad, S., Iqbal, K., Naz, S., Yasmin, S., & Rehman, Z. (2022). A hybrid system for customer churn prediction and retention analysis via supervised learning. Computers, Materials & Continua, 72(3). [5] Li, P., Wei, Y., & Yin, L. (2025). Research on stock price prediction method based on the GAN-LSTM-Attention model. Computers, Materials & Continua, 82(1). [6] Haridasan, V., Muthukumaran, K., & Hariharanath, K. (2023). Arithmetic Optimization with Deep Learning Enabled Churn Prediction Model for Telecommunication Industries. Intelligent Automation & Soft Computing, 35(3). [7] Masturoh, S., Nugraha, F. S., Nurlela, S., Saelan, M. R. R., Saputri, D. U. E., & Nurfalah, R. (2021). Telemarketing Bank Success Prediction Using Multilayer Perceptron (MLP) Algorithm with Resampling. Jurnal PILAR Nusa Mandiri, 17(1), 19–22. [8] Kaisar, S., Rashid, M. M., Chowdhury, A., Shafin, S. S., Kamruzzaman, J., & Diro, A. (2024). Enhancing Telemarketing Success Using Ensemble-Based Online Machine Learning. Big Data Mining and Analytics, 7(2), 294–314. https://doi.org/10.26599/BDMA.2023.9020041 [9] Peter, M., Mofi, H., Likoko, S., Sabas, J., Mbura, R., & Mduma, N. (2022). Predicting Customer Subscription in Bank Telemarketing Campaigns Using Ensemble Learning Models. Machine Learning with Applications. 19. [10] Thabet, H. H., Darwish, S. M., & Ali, G. M. (2022). Measuring the Efficiency of Banks Using High-Performance Ensemble Technique. Neural Computing and Applications. 36. [11] Lew, S. L., & Tang, C. T. W. (2025). Enhancing Segmentation: A Comparative Study of Clustering Methods. IEEE Access. [12] Wang, X.-D., Chen, R.-C., Yan, F., Zeng, Z.-Q., & Hong, C.-Q. (2019). Fast Adaptive K-Means Subspace Clustering for High-Dimensional Data. IEEE Access. [13] Lee, N.-T., Lee, H.-C., Hsin, J., & Fang, S.-H. (2022). Prediction of Customer Behavior Changing via a Hybrid Approach. IEEE Open Journal of the Computer Society. [14] Wang, S., Sun, L., & Yu, Y. (2024). A Dynamic Customer Segmentation Approach by Combining LRFMS and Multivariate Time Series Clustering. Scientific Reports, 14, Article 17491. [15] Wong, M. W., & Su, W. (2025). Segmenting Online Shoppers: A Combined Cluster and Logistic Regression Approach for Forecasting Purchase Behavior. IEEE Access.

Copyright

Copyright © 2025 Dinesh Kumar Katakam, Naga Satya Teja Ventrapragada, Sravan Kumar Mandeti. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.