A Novel Method to Model Urban Ambulance Traffic

Researchers have long-term ambition of improving ambulance operational efficiency with data science

Civilian ridesharing traffic, like Lyft, benefits every day from tailored data-driven routing methods, but what about emergency vehicles? As part of a long-term effort, new research from CDS Moore-Sloan Data Science Fellow Anastasios Noulas and Birkbeck College, University of London researchers Marcus Poulton, David Weston, and George Roussos aims to improve ambulance response times with increased capabilities for data-driven route prediction and optimization. This project relates to research Noulas contributed to last year.

In severe emergency incidents, survival rates plummet after ten minutes of waiting for an ambulance. Major metropolitan areas have the most challenges for ambulances to arrive in this time window. For their project, the researchers focused on London, where ambulance services have a regulatory mandate to reach 75% of patients within eight minutes — in October 2017, only 68% of patients were reached within the mandated time.

To improve response times, the scope of this new project consists of spatio-temporal analysis of ambulance mobility, a multi-layer graph representation of emergency blue lights traffic in London, accurate ambulance speed estimation from GPS data, and deployment of a data-driven predictive ambulance mobility model. Their approach depends on the Blue Lights Road Network (BLRN), which incorporates travel exceptions for emergency vehicles and has five layers based on speed estimations from separate nuanced models.

For their estimations and analyses of ambulance routes, the researchers used the London Ambulance Service’s (LAS) dataset for their entire fleet from March 2014 to December 2016. The dataset includes location data, incident type, and time of activation (departure and arrival). Preliminary observations of the dataset revealed intuitive fluctuations in ambulance traffic depending on time of day and distance from city center. The researchers processed over one million ambulance journeys in their algorithm to construct the BLRN, which uses map-matching techniques to map every traversed road segment from sparse GPS data.

With the detailed road segmentation in the BLRN, the researchers were able to model ambulance movement and predict routes, given a particular incident and initial ambulance location. They compared travel time error and route similarity from their five BLRN layer predictions to the actual LAS dataset. Broadly, their results showed that their predictions often slightly underestimated travel time and that prediction and actual route coincidence usually fell between 70–80%.

In the future, the researchers “aim to explore potential improvements that can be achieved using real-time information as well as traffic and related context information retrieved from external systems.” They emphasize the implications for operational decisions like staffing, deployment, and resource placement. Overall, the researchers have demonstrated the benefits of tailoring travel prediction methods for emergency vehicles which do not abide by civilian traffic patterns.

By Paul Oliver