2025 Men’s NCAA Swimming and Diving Championships
- March 26-29, 2025
- Weyerhaeuser King County Aquatics Center, Federal Way, Washington
- Short Course Yards (25 yards)
- SwimSwam Pick’Em Contest
- SwimSwam Preview Index
- Psych Sheet
Predicting what it will take to score (ie top-8 and top-16 in prelims) in each event at NCAAs can seem a bit like black magic. Events go through periods of strength and weakness, consistency and inconsistency. External factors can have an effect too – remember when the breaststroke events at the Paris Olympics were predicted to be among the fastest in history?
With that in mind, we’ve built a simple model to predict the prelims results for NCAAs this year. A good prediction model should be based on the psych sheets – no matter how an event has been in previous years, how quick it is this year will depend on who’s entered. Year-on-year predictions are great for identifying and analyzing trends, but won’t be the best for individual years.
Our model should be able to answer (among others) this question: given a 16th seed of 19.20 in the 50 freestyle, what does it take to finish 16th and score, and answer just as well if the 16th seed is 19.50?
We’ve adapted the model slightly from that used for the women’s predictions. Time will tell how much better it will be.
When talking about seed position here, we are only including swimmers who finished the race legally. Therefore any swimmers with a DQ or DFS were taken out entirely for that event.
What’s the data, Mr Wolf?
The core data used to make the model is the NCAA results from the past 10 championships (2014-2024). Within that, the model treats every individual position/event combination separately.
Why do we make the model do this? It’s true that by massively increasing the number of datasets, we slash the number of samples in each dataset by the same factor. However, what happens down at seed #30 won’t be the same as what happens up at seed #3. Whilst the aim at both is to try to make a final, there’s almost a necessity to drop time to do so lower down.
If each sample were to include the result for every seed in an event over the 10 years, we wouldn’t be predicting the change from seed time to prelims time for each position – we’d be predicting how the average time at all seed positions changes, and then applying that prediction to every seed.
We’ve built a supervised Machine-Learning regression model using the seeds, seed times, prelims positions and prelims results, which leads to the following predictions for this year’s prelims. We’ll take a look at how well the model does after the championships.
The model does predict the prelims times for every position, but we’re only including the most important in the table here. You can see the full set of predictions (1-50) here
2025 Predictions
| Event | 1st | 2nd | 3rd | 8th | 9th | 16th | 17th | 24th |
| 50 free | 17.80 | 18.23 | 18.39 | 18.63 | 18.64 | 18.83 | 18.84 | 18.99 |
| 100 free | 39.95 | 40.17 | 40.59 | 41.09 | 41.20 | 41.46 | 41.51 | 41.80 |
| 200 free | 1:29.13 | 1:29.86 | 1:30.20 | 1:31.30 | 1:31.50 | 1:31.96 | 1:31.98 | 1:32.30 |
| 500 free | 4:04.51 | 4:07.77 | 4:08.39 | 4:10.64 | 4:11.11 | 4:13.00 | 4:13.12 | 4:14.49 |
| 1650 free | 14:21.25 | 14:25.86 | 14:26.90 | 14:36.09 | 14:37.69 | 14:46.92 | 14:47.51 | 14:51.55 |
| 100 backstroke | 43.67 | 43.86 | 43.99 | 44.81 | 44.95 | 45.07 | 45.09 | 45.29 |
| 200 backstroke | 1:35.96 | 1:36.11 | 1:37.24 | 1:38.57 | 1:38.85 | 1:39.77 | 1:39.88 | 1:39.99 |
| 100 breaststroke | 49.85 | 50.23 | 50.41 | 51.11 | 51.18 | 51.47 | 51.56 | 51.89 |
| 200 breaststroke | 1:49.20 | 1:49.98 | 1:50.17 | 1:50.81 | 1:51.03 | 1:52.46 | 1:52.54 | 1:53.12 |
| 100 fly | 43.30 | 43.68 | 43.91 | 44.43 | 44.48 | 44.83 | 44.83 | 45.09 |
| 200 fly | 1:37.55 | 1:38.43 | 1:39.33 | 1:40.29 | 1:40.51 | 1:41.00 | 1:41.04 | 1:41.66 |
| 200 IM | 1:38.67 | 1:39.28 | 1:39.81 | 1:40.57 | 1:40.74 | 1:41.47 | 1:41.50 | 1:42.35 |
| 400 IM | 3:33.86 | 3:35.24 | 3:37.27 | 3:39.43 | 3:39.55 | 3:41.32 | 3:41.49 | 3:42.58 |
This model is really cool, and the women’s article was neat.
I think all of the 50s and 100s look pretty dead on, which is fun.
Some of the outlier values are funny-if someone goes 1:35 in prelims in the 200 back or 4:04 in the 500 free I question their ability to pace themselves.
If I had to predict an event to be slower across the board than the model says, I’d say the 200 free. Prelims always feels slower than the psych in that event, seems like it always backs up a bit.
If I had to predict an event to be faster than the model says, I’d say the 500 free. I know it’s a risky proposition… Read more »
It will be interesting to see how the model does, what model type specifically are you using? ML is a broad spectrum. SVM? Very cool that you guys are trying something like this.
It would be interesting to run multiple different models and see which ones perform the best. For example, while a neural network may have the highest ceiling, it’s also highly dependent on the data fed into it, and 10 meet’s worth isn’t exactly an Olympic swimming pool-sized amount (I’ll see myself out). Compare that to a simple linear regression, which also benefits from having more data but may not fall apart if it only has, say, 450 entries for a given event.
The SwimSwam writers are all… Read more »
You should use the seed position as an input too rather than splitting it into separate models. As an example, there shouldn’t be much difference between a 15 seed and a 17 seed, and the model could tease out those differences. And it would make your sample sizes significantly bigger.
Adding a feature doesn’t affect “sample size”. Sample size in this case would be the number of entries, whereas adding seed would be considered a “feature” because it’s another point of data from each member of the measured population.
The math for 1st place is scary accurate. I would say the only deviation is the 200BR. I could see multiple guys going under 1:49
100 fly too
200 fly too
Lasco is not going slower than last year.
To be honest, I think he might have some room to go faster in the 100 Back – not but the 200 Back or 200 IM. I do see him making all three A-Finals though