Applying fuzzy integral to performance evaluation in real time strategy game

In real time stategy (RTS) game, there has been much interest in strategy planning research. Past works mostly focused on the development of strategy state matching methods based on weighted summation on attributes. However, large decision spaces with multiple strategy threads (e.g., resource gathering, combat units development) may lead to complicated and even impossible weight calculatious. Many approaches adopted a simplified formula and assuming game features are all independent to each other. Unfortunately, this is not necessary true. We aim to evaluate the strategies´ performance by considering the super-additive properties among variables. In this research, we used Fuzzy Integral to model the feature interactions in RTS games. We also proposed a new fuzzy integral with respect to the RTS game rules called Order-based Fuzzy Integral(Ord-based FI). Compared with the traditional Choquet Integral, our approach demonstrates better learning performance by using real data from Warcraft III game replay logs.