This study investigates how we can effectively predict what type of game a user will choose within the game-based environment iSTART-2. Seventy-seven college students interacted freely with the system for approximately 2 hours. Two models (a baseline and a full model) are compared that include as features the type of games played, previous game achievements (i.e., trophies won, points earned), and actions (i.e., iBucks/points spent, time spent on games, total games played). Using decision tree analyses, the resulting best-performing model indicates that students' choices within game-based environments are not solely driven by their recent achievement. Instead a more holistic view is needed to predict students' choices in complex systems.