par Njoku, Uchechukwu U.F.;Abelló, Alberto;Bilalli, Besim;Bontempi, Gianluca 
Référence Data Engineering for Data Science, Springer Nature, page (81-110)
Publication Publié, 2026-01
Référence Data Engineering for Data Science, Springer Nature, page (81-110)
Publication Publié, 2026-01
Partie d'ouvrage collectif
| Résumé : | Feature selection (FS) is essential in machine learning (ML) to improve model accuracy, interpretability, as well as execution time. Existing FS techniques are categorized across multiple perspectives, including data characteristics, label dependency, search strategies, feature interactions, evaluation methods, and design choices. This categorization provides a structured framework for understanding their strengths, limitations, and practical applications. This chapter provides a compre-hensive overview of FS, covering foundational techniques, a detailed taxonomy, state-of-the-art methods, and emerging trends. The introduced taxonomy is com-pared with existing ones, providing novel insights into the field. Also, an example demonstrating how different FS techniques can be applied to ML is presented, high-lighting their capabilities in addressing ML challenges, including improving model performance, reducing complexity, and lowering training costs. Open problems such as stability under data perturbation, scalability for high-dimensional datasets, and multi-objective optimization are explored, along with practical discussions of FS tools and libraries to bridge theory and practice. In summary, this chapter combines foundational knowledge with future directions, serving as a guide for researchers and practitioners, equipping them to make informed decisions when selecting and implementing FS strategies to address complex real-world ML challenges. |
