Lightweight and Effective Characterization, Screening, and Identification Process for Anomalous Road Traffic Attacks

Smart mapping applications- Google Maps, Apple Maps, Waze- have become ubiquitous in the modern smart city. While the use of these apps undoubtedly eases a number of daily frustrations, their rapid acceptance and integration into every facet of transportation, public safety, and disaster response raises security concerns. Among these concerns is the possibility of an easily-implemented client-side attack; the traffic state shown on a smart map indicates the number of users- not cars- and this disconnect between the ground truth and virtual environment could be exploited by an adversary simply operating a disingenuous number of client devices. In this paper we examine the vulnerabilities exposed by such an adversary and propose a model which combines machine learning traffic prediction and real-time statistical analysis to screen for anomalous traffic attacks. Extensive simulations suggest that our model greatly increases the resources (number of client devices) required to noticeably influence traffic patterns. We show that our model correctly identifies 54.17 % of attacks immediately and reduces the duration of initially undetected attacks by 60.4 % compared to a typical polling system under ideal circumstances, where an adversary has access to an average of 76.0 % more client devices.