We have built our own proprietary model for predicting and calculating drive times.
Process Overview
1 - We take a range of data sets such as OpenStreetMap and country-specific data to build a driving network.
2 - We build a sample of representative routes across the network (covering different trip lengths, road types, surrounding area profiles, etc).
3 - We calculate the driving times for this sample of routes from other leading A to B routing providers, across a 24 hour period (using a midweek day).
4 - We use these sample drive times as part of an iterative machine learning process to build speed profiles (the average driving speed on different types of roads with different features), congestion levels across the course of the day, and other learned parameters (such as turn penalties, stop penalties, etc).
5 - Once the profiles and parameters have been tuned to give results within an acceptably small margin of error to the original sample of drive times, a completely new sample of representative routes is generated, and the accuracy of the drive time model is benchmarked against other providers using the new routes.
6 - If the model passes the final benchmarking exercise, the driving map is released into production.
7 - The models are updated at least every two weeks with fresh data, taking into account any changes in the underlying road network, as well as changes to overall driving speeds.
Accounting for Traffic
The driving model does not use live traffic data
Instead, for each region we sample drive times across an entire 24 hour period (on a weekday) which we then use to model congestion.
Our Custom endpoints provide further flexibility around traffic, with the option to choose between Balanced / Optimistic / Pessimistic traffic conditions
Choosing a different day of the week will not impact on the traffic conditions used, as the time of day is always interpreted as a typical midweek day
Driving Model Accuracy
The extensive benchmarking tests we run our models through ensure that they are highly accurate when compared to other leading routing providers, and much more accurate than open-source routing engines
We publish an open source Benchmarking Tool in our GitHub repo that can be used to run your own benchmarks against other providers