A comparative study of machine learning algorithms for UAV intrusion detection based on cyber-physical data fusion

Introduction The rapid integration of Unmanned Aerial Vehicles (UAVs) into critical sectors has intensified the need for robust Intrusion Detection Systems (IDS). However, traditional frameworks often struggle with the asynchronous nature of heterogeneous data: cyber-only methods lack discriminative power for encrypted traffic, while physical‐only systems are prone to environmental noise. Methods This study proposes an optimized UAV IDS framework leveraging spatio‐temporal feature alignment and cross-domain fusion. The temporal gap between event-triggered network logs and periodic physical telemetry is bridged using a millisecond-level nearest‐neighbor matching algorithm. Six algorithms ‐ Random Forest (RF), K‐Nearest Neighbors (KNN), Support Vector Machine (SVM), K‐means, CNN, and LSTM ‐ are benchmarked on the UAVs‐Dataset‐Under‐Normal‐and‐Cyberattacks across Cyber‐only, Physical‐only, and Aligned Fusion configurations. Robustness is evaluated under 10% Gaussian sensor noise, and interpretability is assessed via SHAP analysis. Results The fusion strategy significantly elevates detection performance, achieving F1‐scores of 0.962 (CNN) and 0.956 (LSTM) under 10% sensor noise, surpassing both the cyber‐only baseline (F1 approx 0.70) and the noise-susceptible physical‐only model. Fusion models maintained superior robustness under noise, whereas single‐domain physical classifiers suffered performance degradation. SHAP interpretability analysis reveals that the model successfully captures cross-domain causalities, specifically identifying stealthy Replay Attacks by detecting inconsistencies between benign‐looking network packets and anomalous battery/altitude fluctuations. Discussion This research establishes a novel paradigm for UAV security by utilizing physical state consistency as an authenticity proof for cyber traffic, offering a resilient defense mechanism for real-world UAV operations.

Authors

Institutions

Publication Details

Journal
Frontiers in Mechanical Engineering
Published
2026-11-19
DOI
https://doi.org/10.3389/fmech.2026.1805899
Primary Topic
Network Security and Intrusion Detection
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

A comparative study of machine learning algorithms for UAV intrusion detection based on cyber-physical data fusion

Haideng Chen, Qingdan Huang, Chuanzheng Wu, Yupeng Ruan
Frontiers in Mechanical Engineering
Network Security and Intrusion Detection
article

A comparative study of machine learning algorithms for UAV intrusion detection based on cyber-physical data fusion

Haideng Chen, Qingdan Huang, Chuanzheng Wu, Yupeng Ruan
article en

Abstract

Introduction The rapid integration of Unmanned Aerial Vehicles (UAVs) into critical sectors has intensified the need for robust Intrusion Detection Systems (IDS). However, traditional frameworks often struggle with the asynchronous nature of heterogeneous data: cyber-only methods lack discriminative power for encrypted traffic, while physical‐only systems are prone to environmental noise. Methods This study proposes an optimized UAV IDS framework leveraging spatio‐temporal feature alignment and cross-domain fusion. The temporal gap between event-triggered network logs and periodic physical telemetry is bridged using a millisecond-level nearest‐neighbor matching algorithm. Six algorithms ‐ Random Forest (RF), K‐Nearest Neighbors (KNN), Support Vector Machine (SVM), K‐means, CNN, and LSTM ‐ are benchmarked on the UAVs‐Dataset‐Under‐Normal‐and‐Cyberattacks across Cyber‐only, Physical‐only, and Aligned Fusion configurations. Robustness is evaluated under 10% Gaussian sensor noise, and interpretability is assessed via SHAP analysis. Results The fusion strategy significantly elevates detection performance, achieving F1‐scores of 0.962 (CNN) and 0.956 (LSTM) under 10% sensor noise, surpassing both the cyber‐only baseline (F1 approx 0.70) and the noise-susceptible physical‐only model. Fusion models maintained superior robustness under noise, whereas single‐domain physical classifiers suffered performance degradation. SHAP interpretability analysis reveals that the model successfully captures cross-domain causalities, specifically identifying stealthy Replay Attacks by detecting inconsistencies between benign‐looking network packets and anomalous battery/altitude fluctuations. Discussion This research establishes a novel paradigm for UAV security by utilizing physical state consistency as an authenticity proof for cyber traffic, offering a resilient defense mechanism for real-world UAV operations.

Frontiers in Mechanical EngineeringVol. 12
China Southern Power Grid (China) (CN)
Network Security and Intrusion Detection
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon