Data-Driven Decision Theory for Player Analysis in Pacman

Computer and videogames have been described using several formal systems - in this paper we consider them as Information Systems. In particular, we use a Decision Theoretic approach to model and analyse off-line, data from Pacman TM players. Our method attempts to calculate the optimal choices available to a player based on key utilities for a given game state. Our hypothesis in this approach is that observing a player's deviation from the optimal choices predicted can reveal their play preferences and skill, and thus form a basic player classifier. The method described builds on work done in (Cowley et al 2006), increasing the scope and sophistication of the model by decreasing reliance on supervision. The downside is a consequent performance hit, which prevents real-time execution of the modelling algorithm. In this paper we outline the basic principle of the Decision Theoretic approach and discuss the results of our evolution toward data-driven classification.