Based on machine learning algorithms, the Google Maps traffic forecasting tool is a fundamental part of the navigation service, as it allows to calculate the fastest route and to tell the user its current time. arrival. To achieve this, the algorithms rely on both historical data and real-time data.
But the coronavirus pandemic has wreaked havoc on the data.
“Since the start of the COVID-19 pandemic, traffic patterns around the world have changed dramatically. We saw a 50% decrease in global traffic when lockdowns began in early 2020. Since then, some parts of the world have gradually reopened, while others maintain restrictions “Google explains in a blog post.
The consequence is that historical data no longer provides a good basis for work. In order to be able to better take into account constantly changing situations, Google was therefore forced to limit its gaze into the past, by “Giving priority to traffic data from the last two or four weeks”. Everything that happened before is now irrelevant.
At the same time, Google has also worked with DeepMind, an Alphabet Group’s artificial intelligence laboratory, to improve predictions of arrival times in some 20 major cities. These arrival times currently have an accuracy rate of over 97%. By concocting Graph Neural Networks (GNN), DeepMind researchers have managed to significantly reduce the remaining margin of error from 16% to 51%.