Feature Selection using Gravitational Search Algorithm in Customer Churn Prediction

Abstract

Customer churn prediction is an essential strategy for companies, especially in telecommunications. Such industries face the challenge that customers frequently switch operators. Due to the higher cost of acquiring new customers compared to retaining existing ones, companies put considerable effort into keeping their current customers. Improving service quality and identifying the point at which customers are likely to terminate their engagement with the company are crucial in retaining customers. Customer Churn Prediction aims to predict potential customer churn by building an effective predictive model. However, the model’s performance is sensitive to unnecessary and irrelevant features. Feature selection is used to eliminate irrelevant features while emphasizing significant ones. This study suggests utilizing a feature selection method to identify significant features and enhance the accuracy of the customer churn prediction model. We propose employing a recently developed evolutionary computation method known as the gravitational search algorithm (GSA) for the feature selection approaches. We elaborate on GSA and the SVM as the classifier to find the optimum features and to improve the prediction accuracy. Our method produced higher precision and AUC scores than the baseline model (without feature selection).