Content area
Abstract
Modern day aircraft are flying computer networks and are vulnerable to ground station flooding, ghost aircraft injection or flooding, aircraft disappearance, virtual trajectory modification or false alarm attack, and aircraft spoofing. Missing aircraft and network attacks are threat to society and the motivation behind the research and the expansion of the body of knowledge beyond the saturated academic research for encryption solution. A rigorous research process answers the research questions by including mixed strategies, data generation method, and data analysis. The questions that arise from the research are:
1) Given known peer accepted ADB-S attacks, what are the indicators of compromise within the ADS-B network traffic using DM/ML techniques?
2) Given the indicators of compromise within the ADS-B network traffic, can ADS-B network attacks be detected at scale using network attack signatures?
The basic design of the study includes simulation and modeling, and data mining techniques. These techniques are used to create a safe environment to exploit the ADS-B protocol and to find indicators of compromise.
The hypothesis is the network attacks on the NextGen ADS-B are detectable using Machine Learning/Data Mining in order to formulate network attack signatures.
The major finding in this research is that by using commodity hardware, there was 80 percent accuracy in predicting the genefrated ADS-B attacks. In conclusion, the answers to the two research questions affirm the hypothesis. The network attacks on the NextGen ADS-B are detectable using Machine Learning/Data Mining in order to formulate network attack signatures. The research indicates that using the features of the machine-learning model (i.e., baroaltitude, velocity, and vertrate), you can characterize flight patterns such that those outside those flight patterns are possible attacks. It took the server 0.36 seconds to preprocess the messages, 11802.38 seconds to fit the model, and 0.36 seconds to apply the model for a prediction. While the fitting took substantial time to finish, the combination of preprocessing and the applying the already fitted model takes less than a second to finish at 0.72 seconds or 720 milliseconds for 20,000 or approximately 27 ADS-B messages every millisecond. The U.S. National Air Space generates 404,058,960,000 message per year or approximately 13 messages per millisecond.
For future work, research on combating the artificial intelligence in where attackers use identify the flight patterns, as done in this research and apply artificial intelligence to produce attacks, while still mimicking the identified flight patterns
