Transit city navigation app seeks to solve bad public transit ETAs with machine learning

Transit, a company that has spent its entire life trying to make it easier to get around cities, is unveiling a new machine-learning powered estimated time of arrival prediction feature to help address one of the most annoying things about taking public transit: not knowing when the next bus is going to arrive.

Montreal-based Transit already offers in-app ETAs for arrivals in the 175 cities where it operates around the world, but the information it provides is based on either real-time data provided directly from a city itself, or crowd-sourced information provided directly by its users. The reality is that neither of these is a necessarily reliable or consistent way to offer truly accurate ETAs.

Of course, predicting when something as fickle as a city bus on city roads filled with human-driven cars will arrive is no easy task. And as Transit Communications Lead Stephen Miller explained on a call, the company also takes into account when determining what constitutes ‘accuracy’ that you’re naturally going to have a much wider margin of error when a bus is 20 minutes away vs. when it’s only a minute or two out.

That said, in Montreal where the new machine learning-generated ETA predictions are launching first, Transit’s team predicts an improvement of around 15% in the accuracy of its predictions for when the next bus operated by Montreal’s transit authority STM will arrive. That should translate into “thousands” of fewer missed buses for commuters, according to the company, and that’s the improvement people should experience on day one, with more improvements to come.

“That’s what we can do with the data that we have today, and the data exploration that we’ve done,” explained Transit’s real-time data lead Juan Borreguero. As we speak we are investigating more ways to improve the data, so for example we’re exploring weather data. We’re exploring analyzing the delays that are present in a certain moment in a city – we call that ‘recency’ – because machine learning is based on historical data, so for example if it’s Sunday and it’s 12 PM we are going to make an ETA based on historical data we have for that time, or around that time, but it can happen that there’s something happening in the neighbourhood at that very second that’s having a huge impact on ETAs, so we’re working on that.”

Borreguero also explained that the company is planning to expand to other markets as well, but that the way transit works in each city where Transit operates varies considerably. Already, the company has gleaned insights about how and why buses are delayed in Montreal vs. how they run in Rome, for instance.

But ultimately, Transit’s entire raison d’être is to improve the lives of people just trying to navigate their cities (and the cities they visit), which is why the company pursued the path of using machine learning to improve on what it’s already been doing to begin with. Relatively fresh off a funding round that includes participation from the venture arms of prominent automakers, Transit looks well-positioned to continue to make life a little easier for public transit commuters.