Automatic dependent surveillance-broadcast (ADS-B) anomalous messages and attack type detection: deep learning-based...

Area of Science:

• Aerospace Engineering
• Cybersecurity
• Computer Science

Background:

• Automatic Dependent Surveillance-Broadcast (ADS-B) is crucial for air traffic control, offering precise GPS-based location data and cost efficiencies.
• ADS-B enhances radar coverage and provides standalone solutions where radar is unavailable.
• The open design of ADS-B lacks inherent security, creating significant vulnerabilities to cyberattacks.

Purpose of the Study:

• To review machine learning and deep learning techniques for ADS-B intrusion detection.
• To develop a detailed attack model for ADS-B, aligning threats with security requirements (confidentiality, integrity, availability, authentication).
• To propose and evaluate an Intrusion Detection System (IDS) using advanced deep learning models for ADS-B message classification and attack detection.

Main Methods:

• Comprehensive literature review of state-of-the-art machine learning and deep learning for ADS-B intrusion detection.
• Development of a categorized attack model detailing potential threats against ADS-B security requirements.
• Implementation and evaluation of three deep learning models: TabNet, Neural Oblivious Decision Ensembles (NODE), and DeepGBM, using accuracy, precision, recall, and F1-score.

Main Results:

• DeepGBM demonstrated the highest performance with 98% accuracy in classifying ADS-B messages and detecting attacks.
• TabNet achieved 92% accuracy, while NODE reached 96% accuracy.
• The study provides a robust comparison of deep learning models for ADS-B security.

Conclusions:

• The proposed IDS effectively classifies ADS-B messages and detects various attack types.
• DeepGBM is identified as a highly effective model for securing ADS-B communications.
• Findings offer critical insights and recommendations for developing future ADS-B security frameworks to mitigate cyber threats.