What Predicts a Hit? I Trained 3 ML Models to Find Out

In many entertainment adaptation decisions, content selections are still instinct-driven. Maybe a producer was vibing with a story or overheard their Gen Alpha nephew mentioning a GOAT title. This subjective approach has often led to expensive missteps and wasted resources for studios when the feature or show turns into a flop. As someone who has worked in the breeding ground of popular webcomics, I asked: what if there was a system that could measure “success potential” of IPs based on real user behavior? Using ML, I wanted to see if I could build a forecasting model that could rank unadapted titles by their predicted commercial success. The Data For my endeavor, I worked with three datasets: Source material metadata of roughly 1,500 titles that included engagement metrics such as views, likes, subscribers, genre, release schedule, and creator usernames Produced show metadata of 1,977 titles including ratings, watcher counts, genre, episode count, and cast Historical webcomic adaptati