García Merino, José CarlosTobarra Abad, María de los LlanosRobles Gómez, AntonioPastor Vargas, RafaelVidal Balboa, Pedro2025-10-032025-10-032025-12-01García-Merino, J.C., Tobarra-Abad, Ll., Robles-Gómez, A., Pastor-Vargas, R., Vidal-Balboa, P. (2025); Título: An Efficient Feature Selection Framework Using Genetic Algorithms; Publicación: 2025 International Conference on Computational Science and Computational Intelligence Proceedings.1865-0929, 1865-0937https://hdl.handle.net/20.500.14468/30316Artículo perteneciente al próximo Congreso CSCI 2025 "International Conference on Computational Science and Computational Intelligence Proceedings" que se celebrará el 3-5 de Diciembre de 2025 en Las Vegas (USA)Proyecto estratégico: Analysis of mobile applications from the perspective of data protection: Cyber-protection and Cyber-risks of citizen information'' y Cátedra Internacional "Smart Rural IoT and Secured Environments''The increasing concern over Internet of Things (IoT) cybersecurity, driven by the growing number of connected devices and their vulnerability to attacks, has intensified the demand for effective Intrusion Detection Systems (IDSs). In recent years Artificial Intelligence (AI) solutions has emerged as a common approach in modern cybersecurity systems, offering powerful tools for detecting complex patterns and threats. However, the limited computing capabilities of IoT environments require models that are not only accurate but also resource-efficient. Feature selection (FS) plays a key role in this context, but traditional techniques such as the Chi-2 test and mutual information focus only on predictive performance, overlooking important practical concerns like inference time and model size. In this work, we propose a novel FS approach based on a custom scoring function that combines accuracy with penalties for inference time and model size. This metric is integrated into a Genetic Algorithm framework to guide the search for optimal feature subsets. The proposed method is evaluated on the NF-ToN-IoT-v2 dataset against standard techniques. Results demonstrate that our approach achieves competitive predictive performance while significantly reducing inference time and model size, highlighting the relevance of considering deployment-oriented metrics in the FS processeninfo:eu-repo/semantics/embargoedAccess1203.17 InformáticaartículoArtificial Intelligence (AI)Intrusion Detection Systems (IDSs)feature selectionInternet of Things (IoT)model efficiency