Customer churn modeling in telecommunication using a novel multi-objective evolutionary clustering-based ensemble...

Area of Science:

• Machine Learning
• Data Mining
• Business Analytics

Background:

• Customer churn prediction is crucial for business growth and cost reduction.
• Ensemble learning models are widely used but face challenges in balancing diversity and performance.
• Developing accurate ensemble models with diverse base classifiers remains a significant hurdle.

Purpose of the Study:

• To propose two novel multi-objective evolutionary ensemble learning models (MOEECs) for enhanced customer churn prediction.
• To introduce a new diversity measure and an objective function to address data imbalance in the second model.
• To evaluate the proposed MOEECs against classical and existing ensemble models using a mobile operator dataset.

Main Methods:

• Development of two multi-objective evolutionary ensemble learning models (MOEEC-1 and MOEEC-2) incorporating a novel diversity measure.
• Implementation of an additional objective function in MOEEC-2 to handle data imbalance and evaluate ensemble performance.
• Empirical evaluation using a mobile operator customer dataset.

Main Results:

• MOEEC-1 achieved 97.30% accuracy and 93.76% AUC, surpassing other models.
• MOEEC-2 attained 96.35% accuracy and 94.89% AUC, demonstrating effectiveness in churn prediction.
• Both MOEEC-1 and MOEEC-2 showed superior performance in accuracy, precision, and F-score compared to previous churn models.

Conclusions:

• The proposed MOEECs offer significant advancements in customer churn prediction accuracy.
• These models outperform existing methods in key performance metrics, highlighting their efficacy.
• The approach demonstrates practical potential for organizational decision-making in customer retention.