Performance prediction at single-action level to a first-person shooter video game

Abstract

Serious games, professional entertainment (e.g. e-sport), or the immersive simulation of critical scenarios by means of virtual reality belong to the scope of the video game industry. With implications in economic (gambling) and professional careers (e-sports), researchers have focused on the high-level analysis of the win/lose chances and on the player’s profile (good/bad performers). At the low-level analysis, the prediction of player’s performance at single-action level, such as in the case of hits in a first-person shooter (FPS) video game, to the best of our knowledge, has not been undertaken yet. In this study we hypothesize that VR video games, embed enough contextual information to predict performance in an FPS game at single-action level. For this purpose, we developed an FPS video game and a single-shoot level prediction model based on virtual world contextual information. Eighteen students of the University of Granada without previous experience in the game played for 45–50 min and generated 600–1200 events each. Every event, which was composed of twenty-contextual-player-centred components of the virtual scenario, were transmitted on-line to a remote server to perform predictions. Data from fifteen out of eighteen participants were used to train the model prediction model. After training, the model predicted “hit”/”miss” with a mean accuracy of 74.1%. In a broad vision, our results suggest that immersive virtual environments bear enough contextual information for accurate predictions even at single-action level. In a closed-loop design, this finding could be used (e.g. in defence, professional e-sports, etc.) to anticipate participant’s actions/decisions before they are taken, and modify the virtual scenario (e.g. abort mission, change environmental conditions) or drive the player (e. g. suggest options, relieve of command, etc.) according to the purpose of the mission.