This project addressed the challenge of effectively deterring threats and mitigating cascading risks at soft targets and crowded places, such as urban transportation systems. Specifically, the project sought to define and quantify threat deterrence strategies across entire transportation systems, not just a specific station or area. Historically, this has not been easy due to the complex, interconnected nature of transportation networks. The proposed solution was to create data-driven models to understand how potential threats and prepared defenses interact across a system. These models intended to pinpoint where security resources are needed most to keep the entire transportation network safe.
The project used (1) Graph Convolutional Networks (GCNs), a type of machine learning architecture leveraging network data, and (2) cascading network failure and robustness simulations that model how different parts of a transportation system—like train stations and rail lines—are connected and how threats might spread through these connections. By studying real-world data from urban rail systems, the developed model intended to predict which stations are most likely to be targeted by attackers, where security is most needed, and how changes in defenses can lower the risk of an attack on the entire network. The project intended to apply its findings in practical scenarios by integrating its algorithms and models into the Virtual Sentry Framework (VSF) command-and-control decision-making system. The resulting insights were to enhance situational awareness, assist in identifying threat hotspots, and provide decision-makers with risk and threat mitigation awareness, resource allocation options, and data-informed guidance for urban transportation networks.
This project improved existing methods by not just looking at individual places that might be targeted by attackers but by examining how an entire transportation system is connected. Current approaches rely heavily on expert opinions, which, though valuable since they are based on local familiarity, experience, and practical knowledge, don’t fully consider how threats or defense actions can affect other separate, possibly distant but interconnected parts of the network. By using advanced machine learning and network simulation techniques leveraging real-world data and information, the project aims to predict how attacks might spread and how changing security measures in one part of the system could impact the rest, making it a more accurate and dynamic way to improve safety across an entire network.
In 2024, the project developed a GCN-based network threat deterrence analysis methodology, assimilated Massachusetts Bay Transportation Authority (MBTA) urban rail network data, created geospatial maps of the subway system and network features, developed a prototype GCN software code, and ran initial network failure and robustness simulations.
The project collaborated with the MBTA to access real-world urban rail network data. The project also engaged with the Cybersecurity and Infrastructure Security Agency (CISA) to understand how network-level threat deterrence metrics may align with national security priorities. These collaborations created access to real-world data and ensure that the project’s threat deterrence metrics apply to practical decision-making for urban transportation security.
