For each year since 1936, Michelin has updated its guidebooks to award or deduct up to 3 stars for each restaurant it inspects. The inspection process is shrouded in secrecy to avoid conflicts of interest and stars are scarcely awarded. The stars have since become one of the most recognized and respected awards a restaurant can achieve, with thousands of chefs from all corners of the globe devoting their lives to the pursuit of just a single star.
In fact, Michelin-starred cooks take the stars so seriously that renowned chefs, such as Gordon Ramsay, have wept over the loss of one or more stars, with some even taking their own lives.
“I started crying when I lost my stars. It's a very emotional thing for any chef. It's like losing a girlfriend. You want her back. I think every top chef in the world, from Alain Ducasse to Guy Savoy, when you lose a star it's like losing the Champions League.”
A single star can have customers flocking to that designated restaurant, bringing about both immense amounts of money and fame. The late and great Joël Robuchon, who himself had accumulated a record 32 Michelin stars, once said that “with one Michelin star, you get about 20 percent more business. Two stars, you do about 40 percent more business, and with three stars, you’ll do about 100 percent more business.”
Customers and investors alike are therefore very keen to predict which restaurants eventually earn a Michelin star before the competition gets fierce.
In the current Michelin-covered export, the imbalance is stark: 61,037 restaurants are treated as No Star after combining Not in Guide, Selected, and Bib Gourmand statuses, compared with 2,706 one-star restaurants, 472 two-star restaurants, and just 138 three-star restaurants.
That makes correctly identifying a restaurant with a star so difficult that a standard ML model would probably be worse off than a basic line of code that constantly outputs a negative Michelin classification. Our model has to find rare signal without turning every expensive, highly rated restaurant into a fake three-star prediction.
Our currently published Project Three Star Model uses a hybrid public rule: raw argmax decides No Star vs starred, then offsets choose the exact starred tier. The numbers in this section are a current-label audit of the Michelin-covered public export, not held-out validation. On that audit, the rule reaches 91.22% exact-tier match rate and 67.08% pooled Macro F1 across the four public tiers. Accuracy is reported for context only; an always-No-Star baseline would score 94.85% accuracy on this same covered export, but only 24.34% Macro F1 because it never recovers a starred restaurant.
| Predicted No Star | Predicted 1 Star | Predicted 2 Stars | Predicted 3 Stars | |
|---|---|---|---|---|
| Actual No Star | 55,953correct no-star calls | 5,054called 1 Star | 14called 2 Stars | 16called 3 Stars |
| Actual 1 Star | 183missed as No Star | 2,492correct 1-star calls | 20called 2 Stars | 11called 3 Stars |
| Actual 2 Stars | 1missed as No Star | 324called 1 Star | 142correct 2-star calls | 5called 3 Stars |
| Actual 3 Stars | 0missed as No Star | 20called 1 Star | 0called 2 Stars | 118correct 3-star calls |
The matrix above shows the tradeoff directly: the hybrid rule keeps the No Star gate much tighter than the offset-only approach while still recovering 94.45% of known starred restaurants. The remaining weakness is precision: many false positives are still 1 Star calls, so the star/no-star decision is better judged with recall, precision, and Macro F1 rather than accuracy alone.
Before choosing the public star-tier setup, we compared raw class probabilities, sigmoid-calibrated variants, class-offset tuning, and a weighted Project Three Star Model + LightGBM + XGBoost probability ensemble. The live public tier now uses raw probability argmax as the No Star gate, then applies the offset tier only inside starred predictions.
We used probability diagnostics to check whether higher scores actually corresponded to higher hit rates before choosing our public exact-tier model.
LightGBM and the weighted ensemble were useful probability references, but the final public setup was chosen only after frozen weights and offsets were retested on fresh city-grouped splits.
Our public exact-tier label comes from the audited prediction export, using raw argmax as the no-star/starred gate and internal offsets only to choose the tier once the raw model says a restaurant is starred.
These are the highest-confidence current non-starred restaurants from our live watchlist inside Michelin-covered target cities. They are not presented as next-cycle guarantees; they are the restaurants the Project Three Star Model most strongly believes deserve a closer look.
| Rank | Restaurant | City | Current status | Prediction | Confidence |
|---|---|---|---|---|---|
| 1 | California Grill | Lake Buena Vista | Not in Michelin Guide | 3 Stars | 91% |
| 2 | Vue | Singapore | Selected Restaurants | 3 Stars | 90% |
| 3 | Flagstaff House | Boulder | Not in Michelin Guide | 3 Stars | 89% |
| 4 | Gary Danko | San Francisco | Selected Restaurants | 3 Stars | 88% |
| 5 | L'Auberge Chez Francois | Great Falls | Not in Michelin Guide | 3 Stars | 84% |
| 6 | Maxim's de Paris | Paris | Not in Michelin Guide | 3 Stars | 81% |
| 7 | Terasa U Zlaté studně | Prague | Not in Michelin Guide | 3 Stars | 79% |
| 8 | Le Club Chasse et Pêche | Montréal | Selected Restaurants | 3 Stars | 77% |
| 9 | Marv & Ben | Copenhagen | Bib Gourmand | 3 Stars | 77% |
| 10 | Prime Steak & Wine | Budapest | Not in Michelin Guide | 3 Stars | 77% |
The research list above stays inside Michelin reporting coverage. The full prediction table also includes supplemental non-covered markets, guide candidates, current Michelin restaurants, and lower-confidence rows.
View full prediction table