Fecha
2025-11-01
Derechos de acceso
info:eu-repo/semantics/openAccess
Título de la revista
ISSN de la revista
Título del volumen
Editorial
IEEE
Resumen
The rapid growth of the Internet of Things (IoT) has intensified the need for effective intrusion detection systems capable of operating in heterogeneous and resource-constrained environments. Deep Learning (DL) has become the dominant approach in recent years, yet the fast pace of research and uneven evaluation practices require up-to-date systematic evidence. This paper presents a PRISMA-based Systematic Literature Review (SLR) of DL-based intrusion and anomaly detection studies published strictly in 2025 (1 January-3 September). Accordingly, the findings characterise the state of the field during the first three quarters of 2025 and should not be read as a comprehensive account of the entire year or of the broader IoT intrusiondetection field. From 1822 screened records, 22 studies were included and analysed across eight research questions covering architectures, datasets, performance metrics, dataset realism, explainability, validation practices, and deployment feasibility. The main contributions of this review are: an updated characterisation of the architectures, datasets, and evaluation practices adopted in the most recent literature within this window; an exploratory pooled synthesis of reported performance by dataset and model family; and a structured assessment of three structural gaps, realism, reproducibility, and deployability, that currently separate the field from practical real-world deployment. Results show that CNN-based and LSTM-based models dominate the 2025 corpus, with evaluations concentrated in semicontrolled testbed datasets. Only one study performed realenvironment testing and two assessed performance on lowresource devices, highlighting limited practical validation. An exploratory pooled synthesis indicated consistently high reported Accuracy and F1-scores in common benchmarks; these pooled values should be interpreted as trend-level indicators of where comparable evidence concentrates, not as a strict performance ranking across models or datasets, and they are obtained almost entirely in controlled experimental settings. Combined with limited reporting of hyperparameters, features, and evaluation protocols, these findings indicate that the field is still at the proofof-concept stage, with substantial gaps in realism, reproducibility, and deployability that must be addressed before practical realworld adoption
Descripción
This is the Accepted Manuscript of an article published by IEEE in "IEEE Internet of Things Journal, 2026", available online: https://doi.org/10.1109/JIOT.2026.3719346
Este es el manuscrito aceptado de un artículo publicado por IEEE en "IEEE Internet of Things Journal, 2026", disponible en línea: https://doi.org/10.1109/JIOT.2026.3719346
Este es el manuscrito aceptado de un artículo publicado por IEEE en "IEEE Internet of Things Journal, 2026", disponible en línea: https://doi.org/10.1109/JIOT.2026.3719346
Categorías UNESCO
Palabras clave
anomalies, Artificial Intelligence, Cybersecurity, Explainable AI, Internet of Things, Intrusions, Neural Network
Citación
E. Fernández-Morales, J. C. García-Merino, L. Tobarra, A. Robles-Gómez and R. Pastor-Vargas, "Deep Learning approaches for intrusion detection in IoT environments: a Systematic Literature Review," in IEEE Internet of Things Journal, doi: 10.1109/JIOT.2026.3719346
Centro
E.T.S. de Ingeniería Informática
Departamento
Sistemas de Comunicación y Control



