Many observers have called this year's race for the Best Director Oscar a toss-up between Richard Linklater ("Boyhood") and Alejandro Gonzalez Inarritu ("Birdman"). But the analytics team at Interpublic's Mithun in Minneapolis said the prize will be Mr. Inarritu's, with 99% odds.

The agency has been picking winners in the four biggest Oscars categories -- best film, best director, best actor and best actress -- mostly for fun over the last 13 years.

But it's also "a demonstration of something real and relevant in the marketing and advertising world right now," spokesman Andrew Hokenson said in an email to Ad Age. "Predictive analytics have become a key driver for marketing.That is, the prevalence of machine learning and predictive analytics, and their ability to drive for marketing."

So how does the agency predict the Oscars? Through readily available data on the internet, it said. "For the most part the data that leads these algorithms is locked away in costly data warehouses. However, we are predicting the Oscars at 89-91% [accuracy] with data that is, for the most part, a couple of clicks away and readily available through a few Google searches," said Mr. Hokenson.

Data used included past award show winners, such as the Golden Globes. For this year's best director prediction, for instance, Mithun's key indicators were Mr. Inarritu's awards as best director at the Director's Guild Awards and the Golden Globes, along with his total Oscar nominations.

Mithun actually built four models, each built on data about films released from 1928 through 1999, and each with varying success rates for the different categories. Mithun later used its predictions and results for 2000 through 2013 to calibrate the models' accuracy, it said.

Last year, one of the shop's four models predicted all four categories correctly. Its other three missed the mark on Best Actress, an award that went to Cate Blanchett for her performance in "Blue Jasmine," though two of the models predicted Amy Adams for "American Hustle" and one chose Meryl Streep for "August: Osage County."

Here are the agency's predictions for this weekend's Academy Awards:

Best Director: Alejandro Gonzalez Inarritu for "Birdman." "Our most accurate model (85.71% accurate over the last 13 years) gives Gonzalez Inarritu a 99% probability to win." The agency noted that past surprises in this category include Roman Polanski's "The Pianist" in 2003, which "stumped all our predictive models that overwhelmingly had picked Rob Marshall for 'Chicago.'"

Best Actor: Eddie Redmayne for "The Theory of Everything." "Our most accurate model (71.43% accurate over the last 13 years) gives Redmayne a 98% probability to win," said the agency, noting that "this model has correctly predicted the last 10 of 11 winners." Past surprises include Denzel Washington in "Training Day" Models had heavily favored Russell Crowe for his role in "A Beautiful Mind."

Best Actress: Julianne Moore in "Still Alice." "Our most accurate model (64.29% accurate over the last 13 years) gives Moore an 87% probability to win," said the agency. "Vegas odds (as of 2/9/15) are giving Moore a 98% probability (perhaps they know something we don't)."

Best Picture: "Birdman." "Our most accurate model (71.43% accurate over the last 13 years) is giving this a 87% probability to win," said the agency, noting that as of Feb. 9, Vegas odds are giving Birdman only a 60% probability.

To get more detail on how Mithun predicted the Oscars, go here.

For the full list of Oscar nominees, go here.