Guiding Feature Selection with Complexity Measures

D. Moctezuma, C. Lancho Martín

Feature selection aims to identify a subset of features that achieves performance comparable to or better than that obtained using the full feature set. The goal is to remove redundant or noisy variables, retaining only the informative ones, thereby enabling the development of more reliable ML models.

This objective is closely related to the concept of data complexity. Complexity measures characterize factors that reflect the intrinsic difficulty of a dataset. In supervised classification, aspects such as class distribution, decision boundary geometry, and noise levels can significantly impact learning performance and complexity measures aim to quantify them.

Although several state-of-the-art studies have applied complexity measures to feature selection with positive results, a comprehensive analysis of how to systematically exploit them is still lacking. In this work, we address this gap by conducting a global analysis and proposing guidelines for their use in feature selection.

Palabras clave: feature selection, complexity measures

GT SW II: Inteligencia Artificial y Aprendizaje Automático
5 de septiembre de 2026  10:00
Aula 21

N. Madrueño Sierro, I. Martín de Diego, A. Fernández Isabel

M. E. Escorihuela Sahún

P. Belén Galipienso, M. J. Nueda Roldán, D. Arcos Limiñana