Deep Learning (DL) for Advanced Persistent Threat (APT) Detection in Cybersecurity
International Journal of Innovations in Science and Technology · 12.02.2026 · 360
DOI: 10.33411/ijist/1764 ↗Kurz gefasst
Fachartikel in der Zeitschrift International Journal of Innovations in Science and Technology, online erschienen am 12.02.2026. Verfasst von Sajjad Ahmed und 5 weiteren Autor:innen. Die Arbeit nennt 66 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).
Abstract (Originalsprache)
Advanced Persistent Threat (APT) is one of the most dangerous types of cyberattacks. These attacks are highly stealthy, long-term, and multi-stage in nature, typically targeting critical infrastructure, businesses, and government organizations. Conventional security solutions and classical machine learning approaches often struggle to detect such threats due to their ability to evade detection over extended periods. Recently, deep learning methods have demonstrated strong potential for APT detection by learning complex temporal and behavioral patterns from large-scale security data. This study presents a comprehensive review and comparative analysis of deep learning–based APT detection techniques reported between 2020 and 2025. The analysis covers the APT attack life cycle, taxonomy of attack types, commonly used benchmark datasets, and the performance of state-of-the-art deep learning architectures applied in modern cybersecurity systems. A quantitative synthesis of the reviewed literature shows that CNN- and LSTM-based baseline models typically achieve detection accuracies between 88% and 93%, with F1-scores ranging from 0.87 to 0.91. In comparison, more recent architectures such as transformer-based models and graph neural networks report mean detection accuracies of 94%–98%, F1-scores between 0.93 and 0.97, and recall rates above 0.92 across multiple benchmark datasets. These models demonstrate performance improvements of approximately 4%–7% in detection accuracy and 5%–8% in F1-score compared with CNN/LSTM baselines, while also achieving relative false-positive reductions in several experimental evaluations. Despite these advancements, important challenges remain, including limited availability of high-quality labeled datasets, difficulties in model interpretability, and constraints related to real-time deployment in operational environments. The study concludes with future research directions emphasizing multi-modal data fusion, explainable AI techniques, online learning frameworks, privacy-preserving detection mechanisms, and scalable deployment strategies to advance robust and practical APT detection systems.
Zitieren
APA
Ahmed, S., Keerio, I. K., Jhatial, M. J., Usman, A., Qutaba, S., & Rasheed, A. (2026). Deep Learning (DL) for Advanced Persistent Threat (APT) Detection in Cybersecurity. International Journal of Innovations in Science and Technology, 360. https://doi.org/10.33411/ijist/1764
Vancouver
Ahmed S, Keerio IK, Jhatial MJ, Usman A, Qutaba S, Rasheed A. Deep Learning (DL) for Advanced Persistent Threat (APT) Detection in Cybersecurity. International Journal of Innovations in Science and Technology. 2026:360. doi:10.33411/ijist/1764
BibTeX
@article{Ahmed2026,
author = {Ahmed, Sajjad and Keerio, Imran Khan and Jhatial, Muhammed Juman and Usman, Anjum and Qutaba, Syed and Rasheed, Abdul},
title = {Deep Learning (DL) for Advanced Persistent Threat (APT) Detection in Cybersecurity},
journal = {International Journal of Innovations in Science and Technology},
year = {2026},
pages = {360},
doi = {10.33411/ijist/1764}
}
EUVIMED