Our race pace prediction estimates drivers' average pace per lap relative to the fastest driver.
Estimates are based on performance in historic races.
Preparing historic lap data
We start by removing non-representative laps (outlaps, laps closely following another driver, laps where the driver was
involved in an incident).
A driver's lap times for a given race are included in the model so long as they complete at least 5 clean laps,
although races with fewer completed laps will have less impact on the final estimate.
Drivers have different strategies during the race and their average pace will be affected by tyre choice, pit strategy, and
fuel load. While all teams will aim to finish the race with the minimal amount of fuel required to pass post-race checks,
drivers who DNF during the race might have slower average pace simply because their laps were completed with higher average fuel load.
We therefore adjust lap times according to projected fuel load at that point in the race, tyre compound and average degradation.
Once these adjustments have been applied to individual laps, we calculate drivers' mean average lap time for the race.
Finally, average times are adjusted for the level of field spread observed in that race to account for circuit length and
track conditions.
Predicting future races
Estimated race pace in previous races is then used to predict race pace in the next one.
Estimates are updated after each weekend, with the size of the update depending on how far the driver under- or over-performed
and the relative uncertainty of the pre-race prediction vs the updated information from the latest race.
As an example, there is more uncertainty around early season predictions soon after
a regulation change and so the update after each race will be large. Similarly, updates will be larger if a driver has recently had a DNF or missed a race.
On the other hand, the estimate for a driver who has performed consistently all season will not move as much after a single race.
What if a driver suffers damage in the race?
If a driver suffers damage in the race or has minimal running time, we downweight that race's information in the update process. Similary, we give less
weight to outlier tracks such as Monaco where gaps are less representative given the difficulty of overtaking.