From Telemetry to Trust: A Lightweight Statistical–Spectral LSTM Framework for UAV Anomaly-Based IDS

Abstract

Unmanned aerial vehicles (UAVs) in mission-critical applications are subjected to advanced cyberattacks of UAV-specific protocols and sensor streams, necessitating intrusion detection systems (IDS) that are sensitive to tight energy constraints and low-latency constraints on airborne edge systems. Current UAV IDS face a critical accuracy–efficiency trade-off: high-accuracy deep learning models exceed the computational capabilities of airborne edge platforms, while lightweight approaches fail to detect sophisticated attacks. This work introduces a novel Long Short-Term Memory (LSTM)-based framework that combines systematic multi-modal feature engineering with compact temporal modeling for resource-constrained UAV deployment. Our solution presents the Windowed Statistical–Spectral (WSS-64) Framework, a multi-modal feature intelligence system that ingests raw UAV telemetry and network traces and runs a 2 Hz compact LSTM model that converts these traces into 64-dimensional frames to provide consistent and accurate detection of classical cyberattacks, including spoofing, false data injection, and denial-of-service malfunctions. Our systematic combination of statistical, spectral, and network characteristics into fixed-dimensional representations optimized for a time model with resource constraints is the main novelty. The method yields a macro-F1 of 0.992 and an AUROC of 0.9995 with 99.21% accuracy and false positive (0.8%) and false negative (0.9%) rates, significantly higher than baseline methods such as Decision Tree (94.85%), Random Forest (96.32%), Support Vector Machine (95.10%), and Convolutional Neural Network (97.45%). The WSS-64 Framework offers O(N) linear complexity for real-time UAV operations.



Received: 22 September 2025 | Revised: 25 November 2025 | Accepted: 6 January 2026



Conflicts of Interest

The authors declare that they have no conflicts of interest to this work.



Data Availability Statement

The data that support the findings of this study are openly available in IEEE DataPort at https://dx.doi.org/10.21227/6f22-py65, reference number [34].



Author Contribution Statement

Hafiz Muhammad Attaullah: Conceptualization, Methodology, Software, Resources, Data curation, Writing – original draft, Writing – review & editing, Visualization. Inam Ullah Khan: Methodology, Resources, Data curation, Writing – review & editing, Visualization. Thabit Mahmood Thabit: Conceptualization, Formal analysis, Resources, Data curation, Writing – original draft, Writing – review & editing, Visualization. Muhammad Mansoor Alam: Validation, Formal analysis, Investigation, Resources, Data curation, Writing – review & editing, Supervision, Project administration. Mazliham Mohd Su'ud: Validation, Formal analysis, Investigation, Resources, Data curation, Writing – review & editing, Supervision, Project administration.