Authors
Andrew Smyth
Carleton Professor of Civil Engineering & Engineering Mechanics, Columbia University
Director of the NSF Engineering Research Center for Smart Streetscapes (CS3)
Co-Chair of Smart Cities Center, Columbia Data Science Institute
Patrick Alrassy
Former PhD student at Columbia University
Senior Data Engineer, Generative AI at Meta, New York
Jinwoo Jang
Former PhD student at Columbia University
Associate Professor of Civil Engineering, Florida Atlantic University
Summary
The Columbia team was sponsored by NYC to develop a comprehensive data analytics approach to leveraging the City’s own DCAS fleet telematics data for the purposes of gaining safety insights into how drivers’ behaviors may be influenced by the road design. Nearly 4000 vehicle metrics such as speed, hard acceleration, hard braking, and trajectories (including left turns) were tracked, and through data analytics approaches, hotspots of key behaviors could be identified. These driver behaviors could be spatially correlated with existing crash databases to learn about linkages. The NYC fleet drivers’ behaviors were compared with the general population of drivers with respect to speed using the Midtown in Motion EZPass-style sensing network to confirm that the set of NYC fleet drivers was a valid proxy for drivers at large. Street segment driver behavior profiles were generated for every street segment across the five boroughs. An innovative technique for map-matching was developed to de-noise the GPS location of data samples from the telematics devices in areas with poor GPS coverage (as is common in NYC’s dense urban canyons and under viaducts). The work was presented on several occasions at the Vision Zero Research on the Road events. The core data sets and software were provided to NYCDOT and installed on local servers to process new data sets as they became available.
Research and Analysis
The research is cited formally in NYC publications as a driver of safety analysis and design:
- Specialized Speed Hump for Bus Routes and larger arterial roads (Speed reducer Program) page 26 mentions:
- DOT continued their partnership with the Smart Cities Center at Columbia University’s Data Science Institute. Academic partners have developed a sophisticated map-matching engine with sophisticated data engineering components to map speed data obtained from DCAS to the New York City street network. DOT has used the maps to help locate sites appropriate for speed cushions, a new type of specialized speed hump suitable for bus routes and larger arterial roads, as part of its data-driven speed reducer program. Columbia University has also started building a tool for planners to access speed data. In order to quickly vet projects and study the connections between hard braking or acceleration and traffic injuries.
- https://www1.nyc.gov/assets/visionzero/downloads/pdf/vision-zero-year-5-report.pdf
- New York City Automated Speed Enforcement Program (The speed profiles this project generated were used in helping to deploy the speed cameras):
- A report published in NYC government DOT website mentioning how the work helped in New York City Automated Speed Enforcement Program
- https://www1.nyc.gov/html/dot/downloads/pdf/speed-camera-report.pdf
- https://www1.nyc.gov/html/dot/html/home/home.shtml
Several peer-reviewed publications have developed:
- Driver behavior indices from large-scale fleet telematics data as surrogate safety measures, Patrick Alrassy, Andrew W Smyth, Jinwoo Jang, 2023/1/1, Accident Analysis & Prevention, Volume 179.
- OBD-Data-Assisted Cost-Based Map-Matching Algorithm for Low-Sampled Telematics Data in Urban Environments, Alrassy, P., Jang, J., Smyth, A.W., IEEE Transactions on Intelligent Transportation Systems, 202.
- A Novel Vehicle Fleet Data-Assisted Map Matching Algorithm for Safety Ranking and Road Classification in Metropolitan Areas using Low-Sampled GPS Trajectories, P Alrassy, J Jang, AW Smyth, Transportation Research Board 98th Annual Meeting Transportation Research Board, 2019.
Related Continuing Applications
This fleet telematics work for NYC was subsequently recognized by the FDNY who worked with the team to improve ambulance response times in NYC using their fleet telematics information coupled with their computerized-aided dispatch (CAD) system data records. This led to changes in the hospital recommendation order provided by the FDNY’s CAD system. These changes saved significant time in to-hospital transport time statistics. That project has led to numerous papers to date:
- NLP-enabled trajectory map-matching in urban road networks using transformer sequence-to-sequence model, S Mohammadi, AW Smyth – arXiv preprint arXiv:2404.12460, 2024
- Bayesian neural networks with physics-aware regularization for probabilistic travel time modeling, Audrey Olivier, Sevin Mohammadi, Andrew Smyth, Matt Adams, Computer-Aided Civil and Infrastructure Engineering, https://doi.org/10.1111/mice.13047, 2023.
- Probabilistic Prediction of Trip Travel Time and Its Variability Using Hierarchical Bayesian Learning, Sevin Mohammadi, Audrey Olivier, Andrew Smyth, 2023/6/1, ASCE-ASME Journal of Risk and Uncertainty in Engineering Systems, Part A: Civil Engineering, vol. 9, issue 2.
- Data analytics for improved closest hospital suggestion for EMS operations in New York City, Audrey Olivier, Matt Adams, Sevin Mohammadi, Andrew Smyth, Kathleen Thomson, Timothy Kepler, Monish Dadlani, 2022/11/1, Sustainable Cities and Society, Volume 86.
- Hospital Load Balancing: A Data-Driven Approach to Optimize Ambulance Transports During the COVID-19 Pandemic in New York City, Edward Dolan, Nicholas Johnson, Timothy Kepler, Henry Lam, Enrique Lelo de Larrea, Sevin Mohammadi, Audrey Olivier, Afsan Quayyum, Elioth Sanabria, Jay Sethuraman, Andrew Smyth, Kathleen Thomson, 2022/4/27, Available at SSRN 4094485.