A machine learning approach to predicting the Oscars

The Academy Awards, or Oscars, are the film industry's highest honour. Since debuting in 1929, they have become a global phenomenon, reflecting their importance in entertainment. Employing AI to predict the Oscars is a relatively underexplored topic. This paper investigates the application of machine learning techniques for predicting Academy Award (Oscar) winners in six major categories: Best Picture, Best Director, and the four Acting awards using a dataset spanning 1961-2025 and incorporating both established and newly engineered features. Logistic Regression proved most effective for Best Picture, Best Director and Male Acting categories, with Support Vector Machines performing best for Female Acting categories, suggesting non-linear patterns in how female performances are evaluated. Key predictors included Directors Guild of America (DGA) wins, which contributed most to both Picture and Director outcomes, and a novel Best Editing nomination feature, which strongly factored into Best Picture success. Results showed top pick accuracy of 85% (Picture), 92% (Director), and 69% (Male and Female Acting), meeting or exceeding benchmarks from prior research while addressing methodological flaws. Importantly, gender-specific analysis and modelling revealed distinct factors influencing male and female acting awards, highlighting unconscious bias in Academy voting patterns.