Performance Analysis Of Machine Learning Algorithms In Detecting Cyber Attacks Targeting Automatic Dependent Surveillance-Broadcast (ADS-B) Systems

In order to support air traffic control services, the U.S. Federal Aviation Administration (FAA) has mandated the use of automatic dependent surveillance-broadcast (ADS-B) in aircraft in certain classes of airspace by January 2020. This system aims to replace the legacy approaches, such as primary and secondary radars, by employing global navigation satellite systems for its operation to generate a precise air picture for air traffic management. The major downside of this system is its security as it broadcasts information of an aircraft such as its position and velocity over an unencrypted datalink. This lack of security makes the ADS-B vulnerable to cybersecurity attacks which can compromise the safety and security of airspace systems. Therefore, it is important to detect these attacks. This dissertation aims at developing methods able to efficiently detect cyber attacks that target ADS-B systems. The proposed methods are based on supervised machine learning models. Therefore, these methods require to be trained using reliable training datasets. In this dissertation, real data as well as simulated one are used to build training datasets and validate the efficiency of the machine learning methods. From this data, several features are extracted depending on the attack type. Results confirm that these methods are reliable, accurate, and independent with high detection and low false alarm probabilities. In addition, unlike existing solutions, these techniques do not rely on information from other surveillance methods and are compatible with current ADS-B systems