Where do speed estimates come from?
Real time speeds
When a Wazer crosses a segment, Waze gets a single data point. A rolling average of these samples (usually over 5 minutes) tells us how traffic is behaving right now. This provides a good understanding of what traffic is doing now, but is a poor predictor of traffic, for example, an hour later.
Historic model
Averages of the observed speeds in past weeks, bucketed by time of day and type of day (weekdays and weekends). Waze blends this with real time data to predict upcoming traffic in the next 30-60 minutes.
Speed limits
When we have no other data, Waze falls back to speed limits, either explicitly added by editors or inferred by road type.
Mixing real time and historical data
Waze uses two guiding principals:
- When things are "normal" give higher weight to history. The more irregular, such as a traffic jam, the more Waze relies on real time data.
- The further down the road, the less Waze relies on real time data.
Traffic history was less reliable during the COVID-19 pandemic. Waze was able to tweak the weighting to rely more on real time data to provide more accurate ETAs.
Types of days
A different traffic model is used depending on the type of day:
- Commute day - A regular day, everyone goes to work and school. Traffic behaves on a weekly pattern - Monday resembles last Monday.
- Summer day - Day with less traffic, such as school holidays.
- Before Waze uses a Summer day traffic model, there needs to be 1 week of consistent Summer-like days before switching to the model.
- Weekend / holiday - Weekend (Sat/Sun for the USA) or community validated holiday, where there is usual no work or school.
- Community defined holidays can be state or country specific.
- Weekends vary around the world. For example, in Israel, the weekend is Fri-Sat. Weekends are defined at a country level.
Information on this page is based on the Waze Communities Office Hours - Pandemic ETA Challenges and Road Width Deep-Dive presentation and follow up post.