An Artificial Intelligence-driven Solution to Promoting Cybersecurity for ADS-B Messages Against Injection Attacks: A Study Conveying Major U.S Commercial Airlines and International Airports

In this study, a solution to detect and classify automatic dependent surveillance-broadcast (ADS-B) message injection cyberattacks is presented. Authentic ADS-B samples from air traffic in proximity to O’Hare International Airport are collected and filtered to eliminate erroneous samples as well as to maintain those only from the five most utilized U.S. commercial airlines. These samples are used to create three simulated message injection cyberattacks: path modification, velocity drift, and ghost injection. Then, both authentic and attack samples are exploited to develop, evaluate, and experimentally validate conventional machine learning detection and classification models. To further evaluate the scalability of such models, authentic and attack samples from ten of the busiest international airports in the U.S are conveyed in their training and development. Resulting evaluation metrics suggest an average validation accuracy (VA), detection rate (DR), misdetection rate (MDR), and false alarm rate (FAR) of 90.9%, 88.5%, 11.5%, and 3.8%, respectively. To improve classification performance, one- and two-dimensional deep learning modeling are proposed while exploring two algorithms for transferring the tabular dataset of authentic and attack samples into images, leading to an average VA greater than 99%, in conjunction with an average DR, MDR, and FAR of 98.9%, 1.1%, and 0.25% respectively. The prediction time of the classifiers is in the millisecond range, suggesting an effective real-time classification solution.