Systematic review of machine and deep learning models for unmanned aerial vehicles cyber threat defense

Unmanned Aerial Vehicles (UAVs) have become integral to defense, logistics, and industrial systems, yet their dependence on wireless links, satellite navigation, and onboard processing exposes them to growing cybersecurity threats. Machine and deep learning methods are increasingly used to safeguard these systems through intelligent intrusion detection, spoofing & jamming identification, and automated threat response. Recent research reveals that convolutional, recurrent, generative, and federated neural models outperform traditional security techniques by detecting complex attack and evolving complex patterns while also supporting rapid, on-board decision making on resource-constrained UAV. Integrating these learning models with blockchain, edge, and fog computing enhances data integrity, reduce latency, and improve coordinated security across UAV networks. Still, practical deployment faces challenges such as vulnerability to adversarial attacks, limited access to standardized datasets, and the opaque nature of deep model decisions. Advances in lightweight model design and attention-based mechanisms are improving real-time performance, yet ethical, regulatory, and privacy issues around autonomous defense remain unresolved. This review synthesis the latest progress in AI-driven UAV cybersecurity and highlights the need for explainable, energy-efficient, and regulation-complaint learning systems to support reliable and secure drone operations in hostile environments.