This module will explore whether (and how) you can predict winning F1 drivers using the DPM Drive Score.
On this Page
Learning Objectives
By the end of this module, you will be able to:
- Locate, clean and use F1 driver data to use for historical analysis and predictive models.
- Use the Drive Score to analyze the performance of an F1 driver.
- Use the Drive to build a predictive model to forecast which drivers will win in future F1 races.
Module Map
This module contains the following progressive lessons.
- Lesson 1: Setting the Baseline: Can Drive Score Predict the New Era of F1?
- Introduces the 'big question' that this module will analyze.
- Lesson 2: The First Checkered Flag: How Did the Model Perform in Australia?
- Learn how and where we source F1 data to use for our analysis and predictive modeling.
- Lesson 3: What are the First Two Races Telling Us?
- We'll review the results of the Drive Score from the first two races and discuss where we're taking things next.
- Lesson 4: Three Races In: Evaluating the Drive Score
- Compare the first four races to see if the Drive Score holds up to our scrutiny.
Published Story
This module was less about a published story and more about how we take our Drive Score for a serious test run. However, you may find it interesting to explore our earlier F1 data story series from the 2025 season, where we explored the races in the latter quarter of the season.
Key Takeaways
We have three key takeaways from this module.
The first is to make sure you test drive any custom or composite metrics you build. Our goal in creating these metrics is not necessarily to take the world by storm; it's more to create a metric where we see a gap for our analyses. We've done this in hockey (with the Team Balance Index), and the Drive Score is similar in that we're wanting to find a composite metric that has broader representation yet can also be used as a predictive variable. No matter what your composite/custom metric, be sure to take it through its paces.
The second is to be open to discarding and trying – through experimentation. No matter what you're trying to do, whether it's analytical, product-related, web-site creation, experimenting to see what works best is of paramount importance. You may find that the metric needs to be simplified, adjusted, whatever; a primary goal to experiment must sit out in front to get the best outcome.
For our F1 model, we needed to strike a balance across pre-season data and in-season data. If the pre-season (or previous season) data is too far away from the current day analysis, then you may find the cars have shifted so much that your predictive model is now redundant. However, too much data from the current season made the data more volatile, so difficult to get crisp predictions. Through three races, we found that the best-performing models are the ones that strike a balance, incorporating real race data without overreacting to it.
Thus, Version 2.0’s 40/40/20 structure currently stands out as the most reliable approach, offering the strongest blend of stability and responsiveness. With Miami up next, we now have a clear direction.