Predictive analysis of real‐time strategy games: A graph mining approach

Abstract

Machine learning and computational intelligence have facilitated the development of recommendation systems for a broad range of domains. Such recommendations are based on contextual information that is explicitly provided or pervasively collected. Recommendation systems often improve decision‐making or increase the efficacy of a task. Real‐time strategy (RTS) video games are not only a popular entertainment medium, they also are an abstraction of many real‐world applications where the aim is to increase your resources and decrease those of your opponent. Using predictive analytics, which examines past examples of success and failure, we can learn how to predict positive outcomes for such scenarios. The goal of our research is to develop an accurate predictive recommendation system for multiplayer strategic games to determine recommendations for moves that a player should, and should not, make and thereby provide a competitive advantage. Herein we compare two techniques, frequent and discriminative subgraph mining, in terms of the error rates associated with their predictions in this context. As proof of concept, we present the results of an experiment that utilizes our strategies for two particular RTS games.