Safeguarding the skies: the rise of machine learning approaches for cyberattack detection in unmanned aerial systems

Abstract Unmanned Aerial Systems (UAS) have become increasingly integral to industries such as agriculture, surveillance, transportation, and entertainment, raising significant concerns about their vulnerability to cyberthreats. Concurrently, cyberattacks have grown drastically in both frequency and complexity. In response, researchers have increasingly turned to Machine Learning (ML) techniques to improve UAS security, though accurately detecting attacks with low computational demands remains a challenge. This Systematic Literature Review (SLR), following PRISMA guidelines, examines ML-based methods for detecting cyberattacks on UAS, analyzing 101 articles published between January 2019 and May 2025. The review summarizes findings on types of feature data, preprocessing techniques, learning approaches, evaluation metrics, and detection designs used in ML-driven cyberattack detection for UAS. Unlike previous reviews, this study addresses the practical implementation of these approaches in UAS. It also highlights key techniques from the literature to help researchers build robust ML models. Furthermore, the SLR outlines the limitations and challenges of ML-based detection methods, including issues related to datasets, evaluation, model complexity, interpretability, and experimental aspects during training and validation. Finally, the review proposes future research directions to advance the field.