The Oscars Forecaster: A Data-Driven Peek at This Year’s Contenders

Just for fun, I built a machine learning model to estimate Oscar nomination and win probabilities across thirteen major categories for the 2025 — 2026 season. Using results from early precursor awards so far, including the Golden Globes, Critics’ Choice Awards, and various guild and society (e.g., Cinematographers, Directors, and Producers) nominations, the model produces probability forecasts rather than simple yes-or-no predictions.

In this post, I’m sharing the nomination probabilities ahead of Wednesday’s Oscar nomination announcement. Win probabilities will follow: preliminary estimates are updated as additional precursor awards conclude, with final winner probabilities calculated once official Oscar nominees are confirmed.

At its core, the model is a two-level logistic regression trained on historical Oscar races and past precursor award results.

Model Performance Interpretation

Since the goal here is to estimate nomination probabilities rather than make simple yes/no calls, AUC is the more informative metric, as it captures how well the model identifies likely nominees overall.

Data Completeness (Current Season)

• Critics’ Choice — awarded
• Actor Awards — upcoming awards
• ASC Awards — upcoming awards
• DGA Awards — upcoming awards
• PGA Awards — upcoming awards
• Golden Globes — awarded
• Oscars (target) — upcoming nominations and awards
• Eddie Awards — upcoming nominations and awards
• BAFTA — upcoming nominations and awards

International Feature

• Training: 2002–2025 (24 years) — 265 historical examples
• Training Metrics: 71% accuracy, 0.67 AUC
• Confidence: Low

Forecast

• The Secret Agent, Brazil, 0.52 (won both the Golden Globes and Critics’ Choice)
• Sentimental Value, Norway, 0.52
• The Voice of Hind Rajab, Tunisia, 0.52
• It Was Just an Accident, France, 0.48
• No Other Choice, South Korea, 0.48
• Sirat, Spain, 0.48

Takeaway: Probabilities are tightly clustered near 0.5 — the field is a close call.

Animated Feature

• Training: 2007–2025 (19 years) — 140 historical examples
• Training Metrics: 61% accuracy, 0.76 AUC
• Confidence: Moderate

Locks

• KPop Demon Hunters, 0.89 (won both the Golden Globes and Critics’ Choice)

Contenders

• Arco, 0.64
• Little Amelie or the Character of Rain, 0.64
• Elio, 0.61
• Zootopia 2, 0.61

Bubbling Up

• Demon Slayer, 0.58
• In Your Dreams, 0.53

Personal Note: KPop Demon Hunters may very well sweep the season. I’d love to see Zootopia 2 put up some fight though.

Original Score

• Training: 2010–2025 (16 years) — 140 historical examples
• Training Metrics: 76% accuracy, 0.87 AUC
• Confidence: High

[embed]

Locks

• One Battle After Another, 0.81

Contenders

• Hamnet, 0.77
• Frankenstein, 0.71
• Sinners, 0.62 (won both the Golden Globes and Critics’ Choice)

Bubbling Up

• F1, 0.51
• Marty Supreme, 0.50

Personal Note: While the data shows uncertainty for Sinners, I think its 2/2 wins prior make a solid case for a locked Oscar nomination.

Cinematography

• Training: 2010–2025 (16 years) — 125 historical examples
• Training Metrics: 79% accuracy, 0.93 AUC
• Confidence: Very High

Forecast

• Frankenstein, 0.76
• Train Dreams, 0.75 (won Critics’ Choice)
• Sinners, 0.74
• F1, 0.67
• One Battle After Another, 0.63
• Hamnet, 0.61

Takeaway: Given high historical ranking power, the top cluster (more than 0.7 probability) is competitive and is sensitive to small changes.

Film Editing

• Training: 2010–2025 (16 years) — 190 historical examples
• Training Metrics: 89% accuracy, 0.96 AUC
• Confidence: Very High

Forecast

• One Battle After Another, 0.38
• F1, 0.34 (won Critics’ Choice)
• Marty Supreme, 0.26
• Sinners, 0.24
• The Perfect Neighbor, 0.10
• A House of Dynamite, 0.10

Takeaway: Despite strong historical fit, the low probability values across candidates suggest an uncertain, highly contested field this year.

Adapted Screenplay

• Training: 2010–2025 (16 years) — 130 historical examples
• Training Metrics: 94% accuracy, 0.98 AUC
• Confidence: Very High

Locks

• One Battle After Another, 0.84 (won the Golden Globes for Writing, and Critics’ Choice)
• Frankenstein, 0.81

Contenders

• Bugonia, 0.75
• Train Dreams, 0.74
• Hamnet, 0.68

Bubbling Up

• No Other Choice, 0.50

Original Screenplay

• Training: 2010–2025 (16 years) — 123 historical examples
• Training Metrics: 91% accuracy, 0.97 AUC
• Confidence: Very High

Locks

• Sentimental Value, 0.93
• Sinners, 0.88 (won Critics’ Choice)

Contenders

• Marty Supreme, 0.77
• Weapons, 0.71
• Jay Kelly, 0.66

Bubbling Up

• It Was Just an Accident, 0.50

Supporting Actor

• Training: 2002–2025 (24 years) — 221 historical examples
• Training Metrics: 86% accuracy, 0.91 AUC
• Confidence: Very High

Locks

• Paul Mescal, Hamnet, 0.89
• Jacob Elordi, Frankenstein, 0.85 (won Critics’ Choice)
• Benicio del Toro, One Battle After Another, 0.82
• Sean Penn, One Battle After Another, 0.82

Contenders

• Miles Caton, Sinners, 0.79 (won Critics’ Choice for Best Young Actor/Actress)

Bubbling Up

• Stellan Skarsgard, Sentimental Value, 0.58 (won the Golden Globes)
• Adam Sandler, Jay Kelly, 0.55

Personal Note: Stellan Skarsgard’s numbers may have been hurt by missing the Actor Awards nominations, along with the entire Sentimental Value ensemble. But his Golden Globe win may propel him toward a very possible Oscar nom.

Supporting Actress

• Training: 2002–2025 (24 years) — 214 historical examples
• Training Metrics: 87% accuracy, 0.94 AUC
• Confidence: Very High

Locks

• Teyana Taylor, One Battle After Another, 0.93 (won the Golden Globes)

Contenders

• Wunmi Mosaku, Sinners, 0.75
• Amy Madigan, Weapons, 0.65 (won Critics’ Choice)

Bubbling Up

• Odessa A’zion, Marty Supreme, 0.54
• Ariana Grande, Wicked: For Good, 0.54

Personal Note: I still can’t believe how competitive Amy Madigan’s campaign is for Weapons — and I mean that in a good, pleasantly surprised way.

Lead Actor

• Training: 2002–2025 (24 years) — 296 historical examples
• Training Metrics: 90% accuracy, 0.96 AUC
• Confidence: Very High

Locks

• Timothee Chalamet, Marty Supreme, 0.81 (won both the Golden Globes and Critics’ Choice)
• Leonardo DiCaprio, One Battle After Another, 0.77
• Michael B. Jordan, Sinners, 0.60

Takeaway: Many candidates are under 0.5 (9 of 12), indicating clearer separation at the top and greater uncertainty in the rest of the field.

Personal Note: I have a feeling this will be a Chalamet sweep.

Lead Actress

• Training: 2002–2025 (24 years) — 301 historical examples
• Training Metrics: 90% accuracy, 0.94 AUC
• Confidence: Very High

Locks

• Chase Infiniti, One Battle After Another, 0.95
• Jessie Buckley, Hamnet, 0.79 (won both the Golden Globes and Critics’ Choice)
• Rose Byrne, If I Had Legs I’d Kick You, 0.74 (won the Golden Globes)

Takeaway: Many candidates are under 0.5 (9 of 12), indicating clearer separation at the top and greater uncertainty in the rest of the field.

Personal Note: This is a Buckley-Byrne race, with Buckley winning.

Directing

• Training: 2010–2025 (16 years) — 154 historical examples
• Training Metrics: 86% accuracy, 0.93 AUC
• Confidence: Very High

Forecast

• Paul Thomas Anderson, One Battle After Another, 0.66 (won both the Golden Globes and Critics’ Choice)
• Ryan Coogler, Sinners, 0.57
• Guillermo del Toro, Frankenstein, 0.57
• Josh Safdie, Marty Supreme, 0.56
• Chloe Zhao, Hamnet, 0.49
• Jafar Panahi, It Was Just an Accident, 0.49
• Joachim Trier, Sentimental Value, 0.48

Takeaway: Probabilities are tightly clustered; no definitive locks are evident.

Picture

• Training: 2010–2025 (16 years) — 564 historical examples
• Training Metrics: 96% accuracy, 0.99 AUC
• Confidence: Near Perfect

Locks

• One Battle After Another, 0.83 (won both the Golden Globes and Critics’ Choice)
• Hamnet, 0.76 (won the Golden Globes)
• Frankenstein, 0.66

Takeaway: Many candidates are under 0.5 (26 of 29), indicating only a small set of films are clearly favored.

Personal Note: I feel that this is a race among three features — One Battle After Another, Hamnet, and Sinners. One Battle After Another may very well sweep the season, with Hamnet very possibly creating an upset.

Note: I received some assistance from generative AI for code debugging, and guidance on this article’s structure and writing.