Negative correlation learning for customer churn prediction: a comparison study.

Area of Science:

• Computer Science
• Data Mining
• Machine Learning

Background:

• Customer retention is crucial for telecommunication companies due to high acquisition costs.
• Accurate prediction of customer churn is essential for effective retention strategies and profitability.
• Existing data mining techniques for churn analysis have limitations in predictive performance.

Purpose of the Study:

• To develop and evaluate an ensemble of Multilayer Perceptrons (MLP) trained with Negative Correlation Learning (NCL) for predicting customer churn.
• To compare the performance of the NCL-based MLP ensemble against traditional flat MLP ensembles and other data mining methods.
• To demonstrate the effectiveness of the proposed model in improving customer churn identification for telecommunication services.

Main Methods:

• Utilized an ensemble of Multilayer Perceptrons (MLP) as the base model.
• Implemented Negative Correlation Learning (NCL) to train the MLP ensemble, aiming to reduce prediction variance.
• Compared the NCL-based MLP ensemble with a standard (flat) MLP ensemble and other common data mining techniques.

Main Results:

• The NCL-based MLP ensemble demonstrated superior generalization performance in predicting customer churn.
• The proposed ensemble achieved a higher churn rate prediction accuracy compared to the flat ensemble.
• Experimental results validated the effectiveness of NCL in enhancing MLP ensemble performance for churn analysis.

Conclusions:

• Negative Correlation Learning significantly improves the predictive accuracy of MLP ensembles for customer churn.
• The NCL-based MLP ensemble offers a more effective solution for telecommunication customer churn identification.
• This approach provides a valuable tool for telecommunication companies to enhance customer retention campaigns and profitability.