Posted by The 94th Academy Awards ceremony is taking place this weekend as the event finally returns to the Dolby Theatre in Los Angeles. As we have done since 2018, we continued collecting the relevant data points about this year’s nominees and updated our Machine Learning models with some tweaks as described later in this post. As a reminder, the dataset is openly available for anyone to clone to their BigML account (get a free one here) and come up with your own predictions if you haven’t done so already.
The Models
We’ve used multiple modeling approaches including OptiML, the popular AutoML feature on BigML that automatically finds the best performing supervised models and individual Deepnets, which powered most of our past Oscars predictions. We also built Fusions, by combining up to the top 10 best performing models from OptiML results per category, e.g., Best Director.
Once our candidate models were created, we made Batch Predictions against the movies produced in 2021 that we had set aside in a separate dataset. Although both OptiML Fusions and Deepnets agreed on most categories, we observed some differences in rankings and relative scores. In the end, we chose to average out the two approaches to come up with our final predictions and rankings.
As usual, given BigML’s emphasis on white-box models with explainability we can dig deeper into models and predictions for added introspection as needed. For example, the partial dependence plot of the above Fusion for Best Costume Design shows how International Cinephile Society and Costume Designer Guild awards (top two predictors for the category) interplay in determining whether a given movie will win the Oscar for Best Costume Design.
The 2022 Predictions
Drum rolls please! Below, we predict the most likely winner along with other nominees in the same category sorted by decreasing scores. These scores are a blend of the OptiML top performing model Fusions and our old faithful Deepnets we’ve been using over the years and they aren’t supposed to add up to 100. In other words, the models are telling us how a movie/artist with a given set of characteristics will do in a given award category based on 20+ years of historical data on that award independent of the other nominees for the same award this year.
Intuitively, Fusion models are likely to balance different techniques to make predictions more robust and in-line with past years as opposed to the individual Deepnets being more likely to pick surprise winners at the expense of missing some favorites. Combining both may be interpreted as a middle-of-the-road trade off that we used to come up with our picks this year.
One more thing, in addition to revealing our model picks, we take note of the leading choices from the BigML Oscars Challenge participants so we can compare and contrast against a mini “wisdom of the crowds” approach.
- Challenge responders are also favoring THE POWER OF THE DOG followed by DUNE and CODA.
- Challenge responders overwhelmingly picked Jane CAMPION too.
- Challenge respondes are taking Jessica CHASTAIN followed by Kristin STEWART just as our models did.
- Challenge responders are going with Will SMITH followed by Andrew GARFIELD.
- Challenge responders are taking Ariana DEBOSE, no surprises!
- The BigML Oscars Challenge responder pick is Troy KOTSUR as well.
- Challenge responders are seeing DON’T LOOK UP and BELFAST on equal grounds. It’s a tie!
- THE POWER OF THE DOG followed by DUNE sums up the BigML Oscars Challenge responder picks.
In summary, it’s quite impressive that our Machine Learning model picks seems to coincide well with the average picks from our community. Of course, we’ll see this weekend whether the joint expectations meet reality.
This year, aside from the high-profile categories above, we have also made predictions for 11 more categories that have to do more with the craftmanship of what takes place behind the cameras. We followed the same approach by blending our Fusions with the individual Deepnets equally for each category and arrived at the following picks. If our predictions hold, with up to 7 wins, DUNE may have a night to remember thanks to its technical prowess.
- Best Cinematography: DUNE
- Best Costume Design: DUNE (NOTE: A virtual toss up between DUNE and CRUELLA)
- Best Film Editing: DUNE (NOTE: OptiML Fusion and Deepnets disagreeing on the ranking)
- Best Sound: DUNE
- Best Visual Effects: DUNE
- Best Makeup and Hairstyling: THE EYES OF TAMMY FAYE
- Best Music, Original Song: NO TIME TO DIE
- Best Music, Original Score: DUNE
- Best Production Design: DUNE
- Best International Feature Film: DRIVE MY CAR
- Best Animated Feature Film: ENCANTO
For what it’s worth, for the more contentious seeming categories with associated notes above, our challenge participants are favoring CRUELLA for Best Costume Design and DUNE for Best Film Editing.
This concludes our 2022 Oscars predictions. Perhaps it’s time for our readers to invite some friends over and impress them with some “educated guesses” on Sunday. After the winners are revealed we’ll follow up with a post-mortem post next week as usual. Good luck to all nominees!
The Official Blog of BigML.com 