Analysis to Improve Customer Retention Using Particle Swarm Optimization and Handling Unbalanced Data in Customer Churn Prediction

This study offers practical insights for developing successful customer retention strategies by highlighting the crucial role that data balancing techniques and advanced machine learning models play in improving churn prediction. This paper uses ensemble machine learning algorithms such as AdaBoost, Gradient Boosting, and XGBoost, along with baseline models like Logistic Regression, Random Forest, K-Nearest Neighbors, Multi-Layer Perceptron, Naive Bayes, and Decision Tree. The publicly available bank and telecom customer retention datasets are used. The Synthetic Minority Oversampling Technique (SMOTE) is applied to address imbalanced dataset issues and improve churn prediction accuracy by integrating machine learning algorithms with Particle Swarm Optimization (PSO) for feature optimization. Significant improvements in handling imbalanced data, reducing bias, and enhancing overall model performance are achieved using SMOTE, SMOTE-Edited Nearest Neighbour (SMOTE-ENN) preprocessing, and PSO-based feature selection. For telecom companies, this study demonstrates the role of machine learning algorithms and PSO in developing effective customer retention strategies and improving churn prediction performance. The paper also examines the impact of PSO for selecting optimal features in combination with SMOTE and SMOTE-ENN for data balancing. The findings show that ensemble approaches, especially XGBoost with PSO, perform better than conventional algorithms.