Customer churn prediction model based on hybrid neural networks.

Area of Science:

• Machine Learning
• Data Science
• Artificial Intelligence

Background:

• Customer retention is vital for organizational sustainability.
• Traditional models struggle with complex nonlinear and time-series data for churn prediction.
• Sample imbalance negatively impacts model performance.

Purpose of the Study:

• To propose a novel hybrid neural network model, CCP-Net, for accurate customer churn prediction.
• To address limitations of traditional machine learning and single deep learning models.
• To enhance customer retention strategies through improved prediction accuracy.

Main Methods:

• Utilized ADASYN sampling for data preprocessing to balance imbalanced datasets.
• Employed a hybrid architecture combining Multi-Head Self-Attention, BiLSTM, and CNN for feature extraction.
• Implemented cross-validation on Telecom, Bank, Insurance, and News datasets.

Main Results:

• CCP-Net demonstrated superior performance across all evaluated metrics compared to existing algorithms.
• Achieved high precision rates: 92.19% (Telecom), 91.96% (Bank), 95.87% (Insurance), and 95.12% (News).
• Showcased performance improvements of 1-3% over other hybrid neural network models.

Conclusions:

• The CCP-Net model effectively enhances the accuracy and robustness of customer churn prediction.
• Its design is suitable for wide application across diverse industries, including finance, telecommunications, and media.
• Provides enterprises with more effective churn management strategies.