Churn prediction in telecommunication industry using kernel Support Vector Machines.

Area of Science:

• Data Science
• Machine Learning
• Business Analytics

Background:

• Customer retention is critical in competitive markets.
• Machine learning models are widely used for customer churn prediction.
• Existing models require optimization for accuracy and feature selection.

Purpose of the Study:

• To develop an advanced customer churn prediction model for a telecom company.
• To evaluate the effectiveness of kernel Support Vector Machines (SVM) for churn prediction.
• To improve prediction accuracy using dimension reduction and data resampling techniques.

Main Methods:

• Kernel Support Vector Machines (SVM) algorithm implementation.
• Dimension reduction using Sequential Forward Selection (SFS) and Sequential Backward Selection (SBS).
• Handling imbalanced data with Synthetic Minority Oversampling Technique Tomek Link (SMOTE Tomek) and Synthetic Minority Oversampling Technique ENN (SMOTE ENN).

Main Results:

• Achieved 99% F1-score and 98.9% accuracy in churn prediction.
• Identified optimal kernel types and important features for SVM models.
• Demonstrated superior performance compared to previous churn prediction approaches.

Conclusions:

• The proposed advanced SVM model significantly enhances customer churn prediction accuracy.
• Dimension reduction and data resampling techniques are crucial for optimizing churn models.
• The findings offer valuable insights for improving customer retention strategies in the telecom industry.