September 28, 2016 12:00 PM - 1:00 PM
We examine human expertise in complex, dynamic, real-time tasks using the video game Tetris. We collected 1 hour of gameplay performance data from 240 players of a variety of skill levels. Our approach is to decompose their task performance into an array of features-- from keypress latency to game-relevant structures-- and identify relevant latent performance variables using principal component analysis (PCA). We then construct principal component regression (PCR) models to examine what behavioral components separates novices, intermediates, and expert players under different naturally-occurring time pressures in the game. To verify these models' efficacy, we use them to predict outcomes of three Tetris tournaments using only a subset of performance data from players' qualifying rounds.