A Hybrid Feature Selection and Soft Voting Ensemble Framework for Accurate Breast Cancer Classification

Abstract

Globally, breast cancer remains a major contributor to mortality among women. This disease can be curable if an early and accurate diagnosis is made. In order to improve the accuracy of predictions, this study suggests a hybrid architecture that combines ensemble classification and optimization-based feature selection. This study employs selection algorithms such as Ant Colony Optimization (ACO), Particle Swarm Optimization (PSO), and Genetic Algorithm (GA) for selecting the most significant variables in the Wisconsin Breast Cancer Dataset (WBCD). Two strategies are then applied to combine the selected features: (1) a majority voting approach, where features chosen by at least two of the three optimizers are retained, and (2) a union strategy that aggregates all features selected by any of the methods. After feature selection, to get the result we have applied classifiers such as Random Forest, XGBoost, Logistic Regression, and a soft voting-based. Our method achieved 98.68% accuracy, with strong F1-score and ROC-AUC values. These results suggest that the ensemble feature selection and classification strategy significantly improves performance compared to individual techniques, making it an encouraging approach for dependable breast cancer diagnosis.