A Performance Evaluation of Machine Learning Algorithms for Breast Cancer Detection Using PSO and CFS Feature Selection

Le résumé fourni par la source

Breast cancer remains one of the foremost global health problems, emphasizing the need for accurate and efficient diagnostic approaches. Machine learning has shown considerable promise in supporting clinical decision-making by enabling timely and reliable disease detection. This study presents a detailed comparison of seventeen machine learning algorithms applied to the Wisconsin Diagnostic Breast Cancer (WDBC) and Breast Cancer Coimbra (BCCD) datasets. The analysis spans traditional, ensemble, and advanced classifiers, with special attention given to two feature selection techniques: Correlation-Based Feature Selection (CFS) and Particle Swarm Optimization (PSO). The research workflow includes data preprocessing, systematic feature selection, and model training using 10-fold cross-validation to ensure robustness and generalizability. Results indicate that ensemble and boosting methods—specifically CatBoost, Random Forest, AdaBoost, and Gradient Boosting—consistently outperform simpler classifiers across both datasets. Incorporating PSO-based feature selection further improves classification accuracy, especially for the WDBC dataset, where CatBoost achieved a maximum accuracy of 98.25%. Although the Coimbra dataset presents greater classification challenges, the use of advanced models and effective feature selection still yields substantial performance improvements. These findings underscore the importance of careful algorithm and feature selection in developing reliable machine learning tools for breast cancer diagnosis, providing valuable insights for the advancement of automated clinical support systems.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.