A Model for Winning Survivor
Summary
The post builds a machine learning model to predict Survivor winners episode-by-episode using logistic regression, with features like time in danger, confessionals, age, number of seasons, and presence of advantages. It compares win and elimination models, discusses training regimes (temporal validation and LOSO), and shows that the model can predict the winner roughly 20-22% of the time—about double random chance—with a dashboard for exploration. It also acknowledges limitations and links the analysis to fan narratives.