Optimizing Ambulances To-Hospital Transports in New York City
Video
Team Information
Team Members
Sevin Mohammadi, PhD Student, Civil Engineering and Engineering Mechanics, Columbia Engineering
Audrey Olivier, Assistant Professor, Department of Civil and Environmental Engineering, University of Southern California
Faculty Advisor: Andrew Smyth, Professor, Civil Engineering and Engineering Mechanics, Columbia Engineering
Abstract
We present data driven techniques to optimize ambulances to-hospital transports through two steps: 1) accurate estimation of ambulance travel times while accounting for uncertainties that arise from traffic conditions, driving behavior, lack of data, etc, and 2) constructing a hospital suggestion pattern based on a probabilistic decision rule. The hospital suggestion pattern ranks the hospitals for every geographical location based on travel time predictions and suggests the closest appropriate hospital if the patient requires hospital care. We propose a data-analytics pipeline that incorporates two sets of data: 1) historical ambulances transport data, and 2) city-wide road segment speed profiles that are constructed by processing city-owned vehicle telematics data. Speed profiles are fed into a network analysis model to estimate average travel times from every atom (FDNY divides the city into small geographical units named atoms) to every hospital. The former data is sparse and has a limited range of travel times and thus, for origin-destination with a small amount of data, the data-based decisions are less reliable, while the latter one is not representative of ambulance travel times. We use a heteroscedastic linear calibration model to calibrate the network analysis travel times using ambulance data to match ambulance behavior. The calibrated travel times are utilized to construct the pattern which is then corrected by ambulance data with high reliability. We use Bayesian bootstrapping to quantify the reliability of ambulance data for each origin-destination pair. The constructed pattern was deployed in the FDNY dispatching system and resulted in an overall decrease in ambulances to-hospital travel times.
Team Lead Contact:
Sevin Mohammadi: sm4894@columbia.edu