Counter attack detection with machine learning from log files of RoboCup simulation

Multi-agent systems have attracted a lot of attention recently. RoboCup Soccer Simulation is treated here as a testbed of such systems. This study aims to facilitate the analysis of team behavior and to clarify the role of different types of team possession in the game results of RoboCup Soccer Simulation. We construct a method of detecting counter attacks, which are one type of team possession, by analyzing the log files of games. To detecting the counter attacks, anisotropy feature and others are introduced. Based on these features, a support vector machine (SVM) based detector was able to achieve a 77% detection rate. The detecting method will be expected to reduce the burden of visually checking log data.