Behind a lot of the work on this blog are a bunch of machine learning models which aim to predict elements of cycling races. I predict whether a race will end in a bunch sprint, whether the morning breakaway will win, which climbs will be most impactful, what the winning power efforts will be over 5 minutes and 40 minutes, which riders have the best chance of finishing highly, and roughly what those win probabilities are by rider. None of this stuff is crazy sophisticated, but the outputs are interesting to look at prior to a race.
With that in mind, I’ve setup a script to collate all of this data for interesting races each week and spit-out automated predictions for a range of models. If this experiment continues, I’ll have the script generate a post each evening and if time permits add a bit of commentary around the data. If not, you’ll just get data and charts.
Race Profile
Alcudia will have a single fairly benign climb in the last third of the race before a downhill drop to the flat finish.
Race Features
The models predict a small chance of the morning breakaway winning – which seems very unlikely to me – and good chance for a bunch sprint (20+ riders finishing together in leading group – which seems low considering last year was a sprint. We’ll see if any riders mix it up on the climb.
The predicted efforts by leading riders are 6.1 watts/kg & 420 absolute watts for 5 minutes and 4.3 watts/kg for 40 minutes. For 5 minute efforts, those are slightly above-average for pro races. For 40 minute effort, that is bang on average.
Race Favorites
These are explicitly generated without looking at betting markets which are very likely are generating better predictions. In particular, the markets seem much more convinced of a sprint.
Girmay won last year and is the favorite again. Presumably he will win in a sprint, though if a smaller group (say 10-20) goes to the line he could easily be in it. The leaders are divided pretty obviously into riders who will win in a sprint (Trentin, Garcia Cortina, Bouhanni, Senechal, Teunissen) and those who will win with a late attack on the climb (Powless, Martin, Konrad).
Historical Power Efforts
Last year’s parcours was very similar with more kilometers. It required fairly benign relative watts (6.2 watts/kg for 5 minutes and 5.4 watts/kg for 20 minutes) and larger absolute watts (430 watts for 5 minutes and 375 watts for 20 minutes). The Z-scores compare to all other pro races where larger watts were required for the longer efforts with fairly high absolute power required for 20 and 40 minutes. This race was fairly high intensity at the end in 2022.
Considering the profile is similar, we can look at where peak efforts occurred for 30 seconds and 2/5/10/20/40 minutes for four groups of riders (top 20 finishers, 21-60, 61+, and non-finishers). The race last year was 174 kilometers with the Sa Batalla climb at 138 kilometers.
Unsurprisingly for a race without much going on the main efforts from 5+ minutes happened on the climb. Top finishers did their strongest 30 second and 2 minute efforts at the end.
The Sa Batalla climb was raced with median 5.5 watts/kg and 371 watts in 2022 race with little difference between those who finished in final group (5.51 watts/kg) and not (5.39 watts/kg).
Science of the Peloton 