Classifying character's action in role-playing game simulation using resilient backpropagation neural network

Abstract

There are many researches using Machine Learning (ML) to create new design contents in computer game. The challenging task is to classify game character’s action using ML because it can be straightforwardly implemented in the game, thereby enhancing character learning about how to deploy strategies under different game situations. This makes the game more exciting. Very Fast Decision Tree (VFDT) can classify character’s actions in computer role-playing game (CRPG) simulation but the accuracy is not much improved when the number of character’s actions is increased. In this research, Resilient Backpropagation (RPROP) can improve such accuracy when the character’s actions increase, so RPROP is implemented to classify character’s action in the CRPG simulation and compared the accuracy with VFDT. The static strategies and the changing strategies are tested in these experiments. The results show that at the high number of training data corresponding to the computer game data, the proposed scheme performs better than the existing method. RPROP can be designed to use in computer game to decrease the complexity of programming script and improve the excitement of the computer game by giving the player more alternatives.