Forward of a grand prix weekend, most of us wish to share predictions or attempt to guess who will come out on prime on a Sunday. Information scientist Mariana Antaya took these chats one stage additional and constructed a machine studying mannequin to attempt to predict F1 race outcomes. Up to now, her mannequin has accurately known as the winners of three grands prix this season.
“I am a very massive Formulation 1 fan,” says Antaya when talking with Motorsport.com. “Machine studying and all these algorithms are actually extensively utilized in Formulation 1 by the groups. I do not suppose as many individuals know, however the race engineers are utilizing this for his or her technique in actual time.
“So, I wished to attempt to predict the winner as a enjoyable train, simply to see, like, how good we will get with the information that is out there.”
To do that, Antaya began constructing a mannequin of her personal. Armed with lap occasions from final 12 months’s Australian Grand Prix, which was sourced from the FastF1 API knowledge retailer, Antaya set about evaluating the 2024 race end result with qualifying performances in 2025.
As soon as the rookies had been faraway from this system, which Antaya admits is the one issue she “interfered with” as there was no knowledge to benchmark towards, she started coaching her mannequin. Utilizing a gradient boosting software, Antaya predicted the lap occasions for the race in Albert Park, and her program accurately picked Lando Norris because the winner.
“I stated on the finish of the video, that is clearly a easy mannequin, and I did not comprehend it was going to foretell proper,” Antaya says. From there, the mission began rising because the F1 neighborhood gathered round to see what number of extra races Antaya may accurately name.
“I wished it to be a crowdsourced sort of factor,” she provides. “So, the entire viewers may say ‘I actually need you to incorporate climate knowledge in it,’ or ‘I actually need you to incorporate the observe classes within the mannequin.’
“I wished individuals to inform me what different options they wished so as to add to the mannequin to enhance it over the course of the season.”
Formulation 1 Fan Mariana Antaya
Photograph by: Mariana Antaya
And enhance it has, because the machine studying mannequin is constant to foretell race winners accurately. This doesn’t imply it’s excellent, nonetheless, and Antaya is now including extra datapoints to this system to assist enhance its accuracy.
“Having extra knowledge goes to assist the mannequin study extra and it is going to have the ability to make higher predictions,” she explains. “In case you solely have a lot knowledge, it will have a really small thoughts, I suppose, and it will not be capable to perceive as a lot.”
With a purpose to increase the thoughts of her mannequin, Antaya added climate knowledge forward of the Japanese Grand Prix, which included the prospect of rain in the course of the race and monitor temperatures at Suzuka. Along with this, wet-weather efficiency of the drivers was additionally added, and this system used this to accurately predict Max Verstappen’s victory on the race.
The subsequent massive step for the mannequin got here forward of the Saudi Arabian Grand Prix this weekend, when it was educated on every crew’s efficiency to this point this 12 months. Antaya defined that the additional strand of information would assist her program perceive that groups like McLaren and Williams have made a step ahead in 2025, whereas others equivalent to Purple Bull aren’t performing constantly in addition to they had been in 2024.
“Now we’re considering extra of a holistic image of how properly the automobile and the crew is performing,” she explains.
‘Shocked’ by the sequence
The sequence of posts on Instagram and TikTok has been rising in recognition with every successive add, and the clips have even reached Formulation 1 itself. A handful of engineers from F1 groups on the grid reportedly reached out to Antaya after she began importing, and he or she’s now trying ahead to discovering out how shut she received to the prediction fashions used within the sequence.
“I have been shocked [by the response]. I have been actually, actually shocked,” she says. “I actually don’t know [how the teams do it]. That is a black field to me, I want I knew. However I hope I am doing it accurately or one thing comparable. They’re utilizing, most likely, far more complicated fashions and far more knowledge that they’ve on the automobile although, for positive.”
Hannah Schmitz, Principal Technique Engineer of Purple Bull Racing
Photograph by: Peter Fox – Getty Pictures
With three out of 5 race winners accurately predicted, Antaya is not resting on her laurels as she hopes to make the predicter much more correct. Forward of the Miami Grand Prix, the information scientist says she needs to start out experimenting with extra complicated machine studying processes to extend the accuracy of her predictions and scale back the imply absolute error of the mannequin, which could be considered the common distinction between the mannequin’s predictions and the race end result.
However whereas the accuracy of the mannequin may enhance due to further datapoints and new processes being applied, Antaya is conscious that in F1 there’ll at all times be unpredictable parts.
“I believe there’s at all times going to be that barrier,” she provides. “It is actually onerous to have the ability to inform that there is going to be a security automobile this lap, and that that is then going to set off another stream of occasions.
“Possibly we may pull previous knowledge on crash share in the course of the race, and that is one thing that we will add as one other function. But it surely’s additionally a sport, so it is not like we will look into the long run and see what is going on to occur on a regular basis.”
On this article
Be the primary to know and subscribe for real-time information electronic mail updates on these subjects
Subscribe to information alerts
