Enhanced Churn Prediction in Telecom with PSO-Based Feature Selection

Customer churn is a significant issue threatening sustainability in the telecommunications sector. This study employs the Particle Swarm Optimization (PSO) algorithm for feature selection to predict customer churn. PSO aims to enhance the performance of machine learning models by identifying the most relevant features in high-dimensional datasets. For this purpose, various data preprocessing techniques, including SMOTE, SMOTENN, undersampling, and oversampling, were applied to the widely used Cell2Cell dataset, and several classification algorithms (Naive Bayes, Logistic Regression, XGBoost, KNN, and Random Forest) were tested. Experimental results demonstrate that feature selection using PSO improves model accuracy and creates simpler, more interpretable models by eliminating redundant features. In particular, a high accuracy rate of 88.8% was achieved when used with the Random Forest algorithm. This study demonstrates that PSO is a powerful feature selection method for customer churn prediction in the telecommunications sector and shows promise for future research.