Machine learning – what is it and how is it relevant for transit?


What is machine learning?

Machine learning can be thought of as a component of artificial intelligence. It looks for patterns in data and uses them to make better, data-backed predictions. Computers use collected data (training data) to make decisions without being explicitly programmed or without assistance from users. The idea is that machines should be able to use the data to learn for themselves, learning from trends or behaviours and using that information to make the best possible decision for future activities. Ideally, over time and with richer data, these prediction algorithms should improve on their own, yielding more precise results. Basically, machine learning uses algorithms to analyze data, learn from it, and then make determinations or predictions about something.1

How is machine learning relevant for transit?

It is no surprise that riders are not keen on spending time waiting around at a bus stop, not knowing exactly when their bus will arrive. Since there are many variables to consider: weather conditions, traffic patterns, change of passenger flow, driver behaviour, etc., it is not always easy to measure real-time travel information. Therefore, prediction algorithms that can calculate travel time are extremely helpful. The abundance of real-time data and innovations in computer technology have led to more accurate algorithms, powered by machine learning.

Precise and real-time travel information in the transit realm is vital for both operators and riders. Not only does it help passengers plan journeys to minimize wait times, it also allows operators to manage fleets. The biggest win? Happy passengers, leading to increased ridership.

TripSpark Technologies has incorporated a new prediction algorithm in its fixed route solutions. Doing so has led to accuracy gains of 30 percent, on average. This has considerably reduced the amount of time riders spend waiting at bus stops every day.

Furthermore, the new prediction algorithm accounts for:

  • Historical Travel Time – if a bus is typically late at the same time on the same day every week, chances are that pattern will continue.
  • Recent Travel Time – if traffic is slow between stops because of a lane reduction, chances are the current trip may be impacted by the same circumstances.
  • Speed Changes – even if a driver leaves the terminal late, it does not mean the bus will be late to arrive at every stop along the route

The graph below illustrates the accuracy gain in predicting bus departure time – (comparing the old algorithm with the new one): 

The ability to predict bus departure times will greatly impact the modern transit world. While the concept of machine learning is not exactly new, it will play a crucial role as we enter the ‘new normal’ of public transit.

About TripSpark:

TripSpark Technologies is a people transportation technology company focused on helping mid-sized public transit agencies and private operators achieve their operational goals. They provide integrated software and hardware solutions for fixed route, paratransit, NEMT, ridesharing and K-12 school.


Add new comment