FPA-Optimized Heterogeneous Ensemble Model for Robust Customer Churn Prediction Across Diverse Industries

Predicting customer churn is critical for sustaining long-term customer relationships, especially in competitive sectors such as telecommunications, banking, and subscription-based services, where retaining customers is significantly more cost-effective than acquiring new ones. This study evaluates several machine-learning approaches for the prediction of churn on three benchmark datasets: BigML Churn, Churn Modeling, and Telecom Churn. Both traditional classifiers, such as logistic regression, random forest, XGBoost, support vector machines (SVMs), and K-nearest neighbors, and ensemble strategies are explored to enhance predictive capability. To further improve ensemble performance, we apply the flower pollination algorithm (FPA) to dynamically optimize the weights of existing base ML classifiers when combined in a heterogeneous ensemble. Each dataset is subjected to a rigorous preprocessing pipeline, including missing value imputation, categorical encoding, feature scaling, and class imbalance correction using the synthetic minority oversampling technique (SMOTE). Model performance is evaluated using accuracy, precision, recall, F1 score, and AUC-ROC. Experimental results demonstrate that the FPA-optimized heterogeneous ensemble consistently outperforms all individual models on the three datasets. For the BigML Churn dataset, the ensemble achieved an accuracy of 98.16% and an AUC-ROC of 99.72%, surpassing the best standalone model. On the Churn Modeling dataset, the ensemble achieved an accuracy of 91.96% and an AUC-ROC of 94.34%, improving upon individual model baselines. The Telecom Churn dataset saw the ensemble deliver 96.05% accuracy, 98.00% AUC-ROC, 95.15% precision, 96.83% recall, and a 95.98% F1 score, highlighting its robustness in identifying churn-prone users. These findings confirm the effectiveness of FPA-optimized heterogeneous ensembles in providing high-accuracy, generalizable churn prediction models. The FPA-optimized ensemble provides a scalable and adaptive framework that can be applied to a variety of industries to support proactive customer retention strategies.