Closing the Bus Prediction Equity Gap
Employing multiple technology interventions to get bus arrival times predictions on par with subway predictions.
The problem
Riders want to know when a bus will arrive so they can plan their lives accordingly.
Transit agencies don't know when each bus will arrive at each stop with certainty, but they can provide continuously updating predictions. There are many challenges to determining accurate predictions:
traffic
the number of passengers boarding/exiting
urban canyons throwing off GPS signals
construction
detours
disabled buses
CAD/AVL (Computer-Aided Dispatch / Automatic Vehicle Location) hardware is critical to a bus network. Radio communications, signage, announcements, operator management, and dispatching all go through it. It is necessarily rugged hardware with expensive installation costs. This means that re-installing or changing out the CAD/AVL system is an unreasonable expectation due to the expense of labor, training, and new hardware.
The projects
Supplemental location sources - GPS devices plus Waze beacons
To boost the accuracy of bus locations without requiring the heavy lift of a CAD/AVL replacement, we pursued additional fleet tracking software that could work alongside the existing hardware. This lower-cost hardware provided both frequent GPS updates and integration with Bluetooth beacons installed in tunnels. This significantly increased the accuracy of our bus location as they move through the city and tunnels.
Silver Line Tunnel Beacons
Once we were clear on the benefits of tying into the Bluetooth beacons, we set out to install the beacons in the Silver Line tunnel to Logan International Airport. We had never been able to offer predictions for these routes as the buses were mostly underground. We worked with our vendors to ensure beacon compatibility and installed beacons throughout the Silver Line tunnels.
Updated prediction engine
With more accurate locations, we submitted a competitive RFP for our prediction engine. We provided all vendors with historical location data to calibrate and then tested them against each other with live data.
The Results
Before our interventions, bus arrival times were ~20% less accurate than subway predictions. This is a clear equity issue as bus riders are far more likely to be transit-dependent. Within 3 months of installing the supplemental location sources, we had closed that gap ~10%. With the new prediction engine, we have achieved near parity.
Affordable interventions like these enable transit agencies to materially improve the rider experience without investing in expensive and highly complex infrastructure redeploys.