An improved Ant Colony Optimization to Uncover Customer Characteristics for Churn Prediction

Customer churn prediction is a critical task in the telecommunication (telecom) industry, where accurate identification of customers at risk of churning plays a vital role in reducing customer attrition. Feature selection (FS) is an integral part in Machine Learning (ML) models which aims to improve performance and reduce computational time (CT). This work optimizes Ant Colony Optimization (ACO) and its structure to empower its capability for customer churn prediction in the telecom industry. The effect of the ACO's hyper-parameters, like the pheromone value, heuristic information, pheromone decay factor, and the number of ants, in the optimization process are investigated. The optimization objective is measured by evaluating the prediction performance of selected features using the k-nearest neighbor classifier. Experiments are performed on three different open-source customer churn prediction datasets. The results are evaluated using several evaluation metrics and compared with three other optimization methods. The findings show that the optimized ACO performs is better than the other comparative methods. The Friedman and Holms test demonstrate that optimized ACO is stable and effective. This work suggests that selected optimal customer characteristics can be utilized to offer valuable insights and reduce churning rate.