Adaptive virtual reality horror games based on Machine learning and player modeling

Fear is a basic human emotion that can be triggered by different situations, which vary from person to person. However, game developers usually design horror games based on a general knowledge about what most players fear, which does not guarantee a satisfying horror experience for every-one. When a horror game aims at intensifying the fear evoked in individual players, having useful information about the fears of the current player is vital to promote more frightening experiences. This work presents a new method to create adaptive virtual reality horror games, which combines player modeling techniques and an adaptive agent-based system that can identify what individual players fear and adapt the content of the game to intensify the fear evoked in players. The main contributions of this work are: (1) a new method to identify individual player's fears using only gameplay data and machine learning techniques; and (2) a new agent-based adaptive game system that can track the horror intensity experienced by players and moderate the use of the horror elements feared by individual players in the game. The results show that the proposed method is capable of correctly identifying players' fears (average accuracy of 79.4% for new players). In addition, results of a user study and statistical significance tests (ANOVA and post-hoc analyses) suggest that our method can intensify the fear evoked in players and positively improve immersion and flow.