A novel hybrid deep learning framework for customer churn prediction using RFM and embedding clustering

Area of Science:

• Data Science
• Machine Learning
• E-commerce Analytics

Background:

• E-commerce customer churn prediction is challenging due to limited labeled data and traditional models' inability to capture complex customer behavior.
• Existing methods often lack effective representation learning or meaningful customer segmentation, hindering accurate churn prediction.

Purpose of the Study:

• To propose a unified hybrid framework for joint customer segmentation and churn prediction in e-commerce.
• To address limitations of existing approaches by integrating RFM-based feature engineering, Deep Embedded Clustering (DEC), and deep learning models.

Main Methods:

• A deep autoencoder learns compact latent representations from transactional data.
• An improved Deep Embedded Clustering (DEC) mechanism segments customers into behaviorally meaningful groups.
• Gated Recurrent Unit (GRU) and Long Short-Term Memory (LSTM) models predict multi-class churn using learned representations.

Main Results:

• LSTM achieved 99.65% accuracy on the Online Retail dataset and 99.83% on the Events dataset.
• GRU achieved 99.77% accuracy on the Online Retail dataset and 99.75% on the Events dataset.
• The hybrid framework demonstrated superior adaptability and performance compared to traditional models like Logistic Regression and Support Vector Machine.

Conclusions:

• Integrating representation learning, clustering, and deep sequential models significantly enhances e-commerce churn prediction performance.
• The proposed framework provides structured, actionable insights for effective customer retention strategies in e-commerce.