Bahrain GP 2022 ?>

Bahrain GP 2022

New season, new cars, new champion……. Hmm, I’m not really sure about the last one since FIA has practically recognized that Masi’s decision in Abu Dhabi GP was wrong by firing him and therefore robbed Lewis his 8th world championship. Nevertheless, the new F1 era with heavily revised cars will bring more opportunities for cars to race closer but also provides teams with a chance to start car design from a blank sheet of paper. That’s why we’ve seen many different approaches during the pre-season testing. I’m not reading too much into the test data and I’ll be waiting for this Saturday to see the pecking order.

The model has not seen any data from the recent testing so I expect all predictions to be off for the first few races. Bottas is predicted to start 2nd, which is highly unlikely after his move to Alfa Romeo. If I ignore the model, I think Max and Sergio will start in the front row while the second one will be occupied by a Mercedes and a Ferrari.

You can access the power predictions here.

Driver Qualifying Prediction Actual Qualifying Result Race Prediction Actual Race Result
Leclerc 6 1 2 1
Sainz 5 3 3 2
Hamilton 1 5 5 3
Russell 13 9 8 4
Magnussen 19 7 10 5
Bottas 2 6 6 6
Ocon 9 11 11 7
Tsunoda 11 16 16 8
Alonso 10 8 7 9
Zhou 15 15 13 10
Schumacher 20 12 19 11
Stroll 16 19 18 12
Albon 18 14 15 13
Ricciardo 12 18 14 14
Norris 8 13 12 15
Latifi 17 20 20 16
Hulkenberg 14 17 17 17
Perez 4 4 4 18
Verstappen 3 2 1 19
Gasly 7 10 9 R
Please share the post:

19 thoughts on “Bahrain GP 2022

  1. Interesting! Though it sounds like these predictions are more for 2021 v2 than for 2022.
    What is the weighting for the 2022 race results once the quali and race results are in? Will the model adjust quickly to the new spec or will it take many races for the new regulation results to start making enough of a dent against the previous years in terms of affecting the predictions?

    1. Yes, the model is basing these predictions entirely on historical results.
      The model’s features track the recent performance of the drivers and the cars, so once we have data for a few races I expect the model to start giving better predictions.

  2. Do you update your predictions after quali? Grid position is one of the strongest predictive features, so it’s interesting to see pre-quali and post-quali predictions. Especially since odds change after quali and there’s some interesting things like Hamilton was given 3/1 odds of podium after quali (and he did make it!) but of course the odds were must worse before quali

    1. Yes, I do update the predictions after the qualifying exactly because the grid position as well as the time difference between the lap times are strong predictors.
      If you see the predictions before the quali, they are not based on the grid position of course. After quali, I do overwrite them with the updated predictions. I don’t keep the pre-quali predictions but I could consider adding the if they are needed.

      1. It would be great to see the probabilities for pre-quali and post-quali the Power Predictor here: https://power.f1-predictor.com/

        I also think that there may actually be better value in betting after quali, but we have to continue monitoring probabilities vs implied probabilities from odds

        1. In the past I had tested whether I would find better value before or after quali, that’s why I stuck with pre-quali predictions.
          If you think of it, it actually makes sense. After quali, the results are way more predictable and therefore you cannot find much value.

          I’m not posting the post-quali probabilities because the power predictions are optimized for the pre-quali odds. Post-quali, I would make different selections but as I said I didn’t find much value to do that.

          1. Good point. It’s a balance between odds available pre vs post quali vs the accuracy of the model with pre-quali vs post-quali results. How did you evaluate the accuracy of your pre-quali model? Simple stuff like avg of correct predictions, area under curve, or something more complex? I suppose the real measure would be the ROI, so ideally you would have built a function to calculate the ROI for each different type of model you used?

          2. The model is predicting pairwise probabilities, i.e. the probability that each driver will finish ahead of each other. Then, I use these pairwise probabilities to get to a final ordering – which is another optimization problem by itself.
            So the model optimized for log-loss vs. the actual historical results. Then, the ordering is trying to optimize some heuristic as described here: https://www.f1-predictor.com/building-an-f1-prediction-engine-predictive-modelling-part-i/
            What do you mean with ROI? Are you referring to the power predictions I guess?

  3. Does the R by Gasly signify that he retired? And if so, why aren’t Verstappen and Perez also marked as retired?

      1. FYI, in the training dataset from Ergast, they would be considered R but with a status of something != 1 (Finished). I found this to be somewhat important, because it helped with accuracy to take “DNFs” as a feature. In simple terms, a driver who is fast, but had high DNFs (either due to mechanical issues w car or accident prone) would be less likely to get a Podium/Points than a driver who is slightly slower but has less DNFs

        1. I’m doing both quali and race predictions before the qualifying. However, I’m updating the race predictions after qualifying based on the quali results.

        1. I published the Saudi Arabia predictions just now. Feel free to subscribe to the newsletter to get informed whenever I publish something new.

    1. The model does not take into account the practice session data. So, I’m doing the predictions before the practice. Still, I’m updating the race predictions after qualifying based on the quali results.

Leave a Reply to Anonymous Cancel reply

Your email address will not be published.