Group items tagged

Abstract: "Video games that customize to a player's experience level and abilities have the potential to allow a broader range of players to become engaged and maintain interest as they progress in experience level. A game that uniquely customizes the player's experience could attract additional demographics to gaming, which will result in a distinct edge in marketability and potential revenue. This thesis examines a subsection of adaptive gaming systems from the perspective of identifying game factors that alter the level of difficulty. Our focus is to provide a solution useful to both research and commercial gaming communities by developing a system that simulates results offline yet can be integrated into online play. While online performance is the main goal of an adaptive system, the offline simulation provides several benefits. Offline simulation allows the elimination of insignificant factors from inclusion in the training and evolution phase of machine learning algorithms. In addition it provides commercial games with a useful tool or method for performing game balancing and level tuning. To test our approach we designed a test-bed version of the game Pac-Man. The experimental testbed alters environment variables to evaluate their effect on a set of selected response variables. Observing the results of several response variables provides the potential to represent multiple player states, though our focus is on controlling the difficulty for a player. The testbed will simulate the actions of both Pac-Man and the ghosts over a variety of different settings and strategies. The evaluation of a factor's significance and its effect size are calculated using a factorial analysis approach. This method allows the identification of factors relevant to both individual strategies, and the set of all player strategies. Finally, as a proof of concept for both the online and adaptation prospects of this method, we developed a prototype adaptive system. Utilizing the releva