A novel customer churn prediction model for the telecommunication industry using data transformation methods and...

Area of Science:

• Data Science
• Machine Learning
• Telecommunications Industry

Background:

• Customer churn remains a critical challenge in the telecommunication industry.
• Existing customer churn prediction models show room for performance improvement.
• Leveraging customer relationship management (CRM) data is key for churn analysis.

Purpose of the Study:

• To improve the accuracy of customer churn prediction in the telecommunication industry.
• To investigate the impact of data transformation and feature selection on model performance.
• To optimize machine learning models for better churn identification.

Main Methods:

• Employed machine learning models for customer churn prediction.
• Utilized univariate feature selection techniques.
• Optimized hyperparameters using the grid search method.
• Applied data transformation methods to enhance model training.

Main Results:

• The proposed technique demonstrated significant improvements in prediction performance.
• Achieved up to 26.2% increase in AUC (Area Under the Curve).
• Showcased a 17% improvement in F-measure for churn prediction.

Conclusions:

• Data transformation and feature selection are crucial for optimizing customer churn prediction models.
• The investigated methods offer a substantial enhancement in predictive accuracy for the telecommunication sector.
• The findings provide a pathway to more effective customer retention strategies.