Evolutionary Approach to Balance Problem of On-Line Action Role-Playing Game

We address the balance problem of on-line Action Role-playing Games(ARPGs) by using an evolutionary architecture which includes intergration with an improved Probabilistic Incremental Program Evolution(PIPE) and Cooperative Coevolutionary Algorithm(CCEA). We construct a game model of the basic on-line ARPG and use it to obtain the training data for a statistical model named Linear Logistic Regression Model(LLRM) by monte carlo method, then we calculate fitness value of the individual based on the prediction of the LLRM. Moreover, some experiments are made to demonstrate the evolutionary architecture is able to find a set of well-balanced Ability-increasing Functions(AIFs), which determine how to increase the ability value with level, for the target game more efficiently and logically.