An Anomalous Delay Detection System for the New York City Metro

Abstract

The New York City metro is one of the most vital and expansive transportation networks in the world. Its aging infrastructure, however, makes it highly vulnerable to operational delays. Currently, mitigating system-wide disruptions relies heavily on institutional knowledge rather than dynamic computational network analysis. This thesis addresses the problem of quantitatively measuring the network's resilience by developing an anomalous delay detection system. By parsing static General Transit Feed Specification (GTFS) data, the subway system is modeled as a directed graph where physical stations serve as nodes and train routes as directed edges. This static baseline is then merged with real-time GTFS arrival data through a custom processing pipeline, utilizing approximate string matching to deal with formatting noise and overlapping time standards. Statistical thresholds derived from historical delay variance are applied to differentiate normal operational noise from severe, systemic disruptions. Ultimately, this project establishes both a comparative data analysis repository and an interactive graphical user interface. The analytical repository contrasts distinct operational realities, specifically quantifying the structural shifts between weekday commuter strain and weekend infrastructure maintenance. The GUI serves as a minimum viable product, translating these complex, system-wide stress states into human-readable map visualizations and critical delay reports. Together, this framework provides a robust computational foundation for dynamically tracking network hotspots and identifying structural transit vulnerabilities.