Since 2014, there have been seven different riders to win the 450 Supercross Championship. There has not been a repeat winner since Ryan Dungey's three-peat from 2015 to 2017. Just last year, Jett Lawrence took the industry by storm. Not only did he come off one of the most exceptional 250 careers to date, but he also began his 450 career by winning his rookie Motocross, SMX, and Supercross championships.
Over the years, the tracks have gotten faster and the competition deeper, but the keys to winning it all have remained the same. Or have they? We decided to simulate the entire 2025 Supercross season hundreds of times to not only predict the outcome of the 2025 Supercross season but also to understand what is most important to winning a championship. We analyzed everything from qualifying positions to lap times to consistency to determine what metrics build Supercross champions.
We decided to simulate the entire 2025 Supercross season hundreds of times.
We're not Vegas sports bettors peering into a crystal ball to predict the outcome of the 2025 season. But with modern technology, we can use historical data and advanced AI models to forecast what might happen. Our simulation leverages over a decade of racing stats to predict rider performance and outcomes for the top 30 riders heading into the 450 championship. Make sure to follow along with our Moto Metrics for the entire 2025 racing season where we will be constantly updating our championship odds as the season progresses.
How It Works:
At its core, the simulation looks at how riders have performed in the past, analyzing details like lap times, race finishes, consistency, and even who grabs the holeshot. Using these stats, it forecasts how these riders might perform next season. We factor in trends over the career of the rider but also add a bit of random variation based on how consistent the rider is, using LITPro's consistency scoring metric. We run this forecast hundreds of times to account for the unpredictable nature of racing. Then, each simulation is fed into an advanced neural network to predict the probability of winning the championship. Here's how it works:
-
Data, Data, and More Data: First, we pull the historical data for every rider and also compute new metrics. All the data is collected from AMA live timing and scoring. We measure how often they set the fastest lap, the percentage of times in the top 5, podium finishes, wins, etc. We also include consistency scoring on a per-race basis.
-
Forecasting Next Season Stats: Then, we use advanced techniques, like ARIMA (a statistical forecasting tool), to predict these key stats for each rider. We took 17 stats for each rider, which include things like how many laps they might lead next year, podium percentage, and even predicting the number of race wins.
-
Predicting Unknown: Supercross is as much about chaos as it is about speed. Injuries, mechanical issues, or a rider simply hitting their peak can drastically change the outcome of a season. To account for these unknowns, we simulate the 2025 season hundreds of times. Each simulation incorporates random variations (based on the riders lifetime Consistency Score) to mirror the unpredictable nature of racing.
-
Simulating the Season: The results from these simulations are passed into a neural network, a type of AI designed to learn complex patterns. This network evaluates the simulated data and predicts the probability of each rider winning the championship based on over a decade of past supercross racing.
- Ranking the Riders: At the end of the simulation period, the model generates a leaderboard with championship odds for the top 15 riders. These odds are normalized using advanced techniques to provide a clear picture of the competitive landscape. We compile the results to not only calculate the probability of winning the championship but also to predict the number of wins each rider could have.
Why This is Cool:
Supercross drama thrives on the thrill of the unknown. Every year, unpredictable events happen: who wins the opener, which riders get injured, and of course who will emerge as the champion. While this simulation doesn't remove the excitement of unpredictability, it offers a fresh way to appreciate the sport that many other racing industries are utilizing. By breaking down performances at this level, we can better understand the dynamics of each race and the season as a whole. Not only that, this article plus our entire moto metrics series will give you more stats to talk and bench race with than Weege or RC will ever dive into on the live broadcast.
Are holeshots more important than lap consistency? Does setting the fastest lap time matter more than top-five finishes?
The model doesn't just forecast winners for betting odds; it offers insights into what makes each rider successful. In other words, we know exactly what it takes to build a Supercross champion. Execution, on the other hand, is an entirely different story. By analyzing feature importance (how much each stat contributes to a championship), we can identify traits that separate champions from the rest of the field. Are holeshots more important than lap consistency? Does setting the fastest lap time matter more than top-five finishes? These findings offer valuable insights for riders of all levels and fans alike. So, let's jump into the results of our simulation.
How Championships are Built:
We have seen through the decades that winning championships requires more than just raw speed. The riders at this echelon have proven that. It also takes more than winning races. But how can we uncover which statistics are most important in determining a champion? Using insights from random forest models to "detangle" our neural network, we have gathered which statistics riders should focus on throughout the 17 rounds of Supercross. The table below highlights the overall importance of each metric we fed into our simulation.
| Statistic | Overall Importance |
|---|---|
| Podium | 37.7% |
| Wins | 20.8% |
| Top Five | 17.3% |
| Finishing Race | 11.5% |
| Lowest Avg Lap Time | 6.7% |
| Set Fastest Lap | 2.8% |
| Laps Led | 2.4% |
| Starting Pos. | 0.3% |
| Qualifying Pos. | 0.1% |
| Lap Delta | 0.1% |
| Holeshots | 0.2% |
| Consistency Score | 0.1% |
The single most important factor in determining championship success is podium finishes. Nothing else even comes close. Not only was it the most important stat across hundreds of simulations, but finishing consistently on the podium is nearly twice as valuable as winning races. In fact, winning was never the most important metric to championship success. So while winning races does net the most points, championships are often decided by a balance of race wins and steady podium finishes. The historical data backs this up too, just ask Ryan Dungey.
Winning was never the most important metric to championship success.
Even more surprising is that managing to consistently finish inside the top five is nearly as important as winning the race. Heck, even just making the main event and finishing the race guarantees high placement in the championship, particularly for privateer efforts. We have seen time and time again that the riders who manage to make nearly all main events finish inside the top 15 in points.
What isn't surprising is that speed doesn't determine champions. In fact, sometimes it's the opposite. The neural network quickly picked up on this. Only 45% of champions over the past decade could claim that they were, on average, setting the fastest lap times in most of the races.
What isn't surprising is that speed doesn't determine champions. In fact, sometimes it's the opposite
Predicting the Champion:
After the simulated dust settles, we can finally determine which rider is most likely to win the 2025 season. Following each simulation, the riders are given a likelihood of winning the championship. We think this year is going to be incredibly close. By combining all the simulations, we have the predicted chance that each rider will win the championship showcased in the table below.
| Rider | Probability to Win Championship |
|---|---|
| J. Lawrence | 12.5% |
| E. Tomac | 12.2% |
| C. Webb | 12.1% |
| C. Sexton | 10.8% |
| K. Roczen | 10.2% |
| H. Lawrence | 9.4% |
| J. Anderson | 6.8% |
| J. Prado | 4.2% |
| A. Plessinger | 3.7% |
| J. Cooper | 3.1% |
| C. Craig | 3.0% |
| D. Ferrandis | 3.0% |
| C. Nichols | 3.0% |
| M. Stewart | 3.0% |
| S. McElrath | 2.0% |
| J. Barcia | 1.0% |
The above table might give overall odds to win the championship, but after simulating hundreds of championships, a clear favorite shined through it all. The graphic below showcases how often each rider finished in each championship position per simulation. So, while Jett, Eli, and Cooper all have, on average, an equal likelihood to win a championship next year, it was Jett who actually won 80% of the time.
So, while Jett, Eli, and Cooper all have, on average, an equal likelihood to win a championship next year, it was Jett who actually won 80% of the time.
Jett's stats back this up too. In his last three Supercross titles (including 250s), Jett has been dominant. In the 250s, he managed to finish on the podium in every race for two seasons in a row. Last year, he also ranked first in podium percentage among all 450 riders. As we previously discussed, podiums win championships. And while Jett may be chasing McGrath's win record, Team HRC has clearly found the formula to winning championships is through podium finishes.
But How's the Accuracy?
Just to make sure that we aren't going off-season stir crazy, we decided to test our model to see just how accurate it really is. This is actually a simple task as we repeat the simulation but remove everything on the 2024 season. Doing so, yielded fantastic results. Not only did our model predict the correct championship winner, it also picked the correct top 5 (just not in the right order). Check out the table below.
| Rider | Model Prediction on 2024 Season | Actual Finish |
|---|---|---|
| Jett Lawrence | 1 | 1 |
| Eli Tomac | 2 | 4 (-2) |
| Chase Sexton | 3 | 3 |
| Cooper Webb | 4 | 2 (+2) |
| Jason Anderson | 5 | 5 |
| Ken Roczen | 6 | 7 (-1) |
| Aaron Plessinger | 7 | 11 (-4) |
| Justin Cooper | 8 | 6 (+2) |
| Justin Barcia | 9 | 8 (+1) |
| Colt Nichols | 10 | 20 (-10) |
Our simulation process was not only able to predict Jett winning his rookie 450 season, we also correctly placed 8 of the top 10 riders, 3 of which were picked correctly. Eli Tomac, Aaron Plessinger, and Colt Nichols all missed at least two rounds last season which the model cannot account for. We are quite happy with our results in re-creating the 2024 season which builds confident in our 2025 predictions above.
That's Not All:
Finally, since we have gone through the trouble of simulating the season hundreds of times, we are also calculating how many wins each rider may have in the 2025 season. In the table below, we've highlighted win ranges, showcasing the floor and ceiling for each rider, as well as how many we expect them to have. While the expected wins add up to 17, the minimum and maximum wins showcase the potential of other riders not on this list winning, or one rider dominating the season.
| Rider | Minimum Wins | Expected Wins | Maximum Wins |
|---|---|---|---|
| J. Lawrence | 2 | 5 | 8 |
| E. Tomac | 2 | 4 | 7 |
| C. Webb | 3 | 3 | 9 |
| C. Sexton | 1 | 2 | 6 |
| H. Lawrence | 1 | 2 | 4 |
| K. Roczen | 1 | 1 | 4 |
The table above is the result of hundreds of simulations. In some of the individual simulations, riders like Plessinger, Craig, Anderson, Prado, McElrath, Cooper, and Ferrandis all found the top step of the podium on rare occasions. However, given how stacked this year seems to be, only the riders in the table are expected to win.
While we were at it, we also broke down what stats are most important for winning. Unsurprising, given the tight and technical nature of Supercross, starts are key to winning races. Over the last five Supercross seasons here are some quick stats on making it to the front of the pack all based on starting position:
- No rider won a race that finished the first lap 9th or worse
- 65% of riders who got a holeshot finished on the podium
- 75% of riders who got a holeshot finished in the top 5
- Of those that won a race, 47% were in first by the end of the first lap; 30% nailed the holeshot.
You can learn more about perfecting your starts using LITPro's Gatedrop Analytics. Read more here!
4 comments
Phil – I did the best with Prado without excluding him entirely. I based his rookie season stats using a baseline of other similar European riders first 450 season like Musquin and Ferrandis. Plus we have some stats on him last year, but I think we all expect him to improve. Not perfect, but this is why I tried to find where he’s most likely to place at season end. I think a range of 7-12th centered around 10th would be respectable / expected. And yes, the 250 class does throw a wrench into things. For simulating stats, it does consider their 250 career but the neural network is only trained on 450 racing data. This is something that we could definitely tweak. Modeling can easily dive into an iteration loop, and it can be difficult in understanding where to stop and say “I’m happy with this”. And yep, there are things we can’t account for like team/bike changes, but glad you found it all interesting and thank-you for the comment!
Greg – Yes, I plan to update championship odds based on how the season is progressing starting with the first round. Plan on doing the same with 250s once we get the which rider / which coast thing figured out.
Jozo – I’m not sure how much awareness the teams have, and have never had the opportunity to communicate with any of them about it. That would make for a great conversation though, and I think fans would be interested in understanding how a team approaches a championship mindset. For example, how do they approach the first round, deal with setbacks, etc. Right now, we don’t have anything on when a rider peaks during the season. Even though the entire season is simulated, it’s simulated all at once and not on a round-by-round basis. The model knows there are 17 rounds, but doesn’t particularly look at them one after another. This approach has been something we have discussed though. As we know, some riders perform at certain venues well above their average (Tomac and Daytona, Barcia at A1, etc). Definitely something we can build on as we continue to tune this model and provide more data insights.
I wonder how aware team/team managers are of these season winning metrics? Do they manage more by gut feeling or by the numbers? I even wonder if a metric could be developed and what the predictive power is of when the rider peaks during the season. I’d guess peaking later would show up in there somewhere.
Will you be using this formula for each race or just for the overall 2025 season ???
Very interesting. Two things stand out to me:
1. How high Prado’s championship probability is, which I would assume to be purely due to the low number of data points off of which to forecast.
2. That 250 class results have a fairly high influence on this model’s outcome as illustrated by Nicols’ and McElrath’s championship probability being higher than Barcia.
The data doesn’t lie, but there are definitely factors that it can’t account for, like team/bike changes for the coming year. But it’s a great sanity check if you are serious about fantasy/betting.