Abstract

The goal of our research is to analyze the stress level of a human player during a game session. In this research, we propose a system to classify certain human states recorded by EEG analysis during playing computer games activity. The classification procedure is off-line, thus the recording will not interfere the game. From that recording, we can differentiate certain player states. The games we used for the experiment are different challenges in racing games, chess, and first person shooter with different types of difficulty levels.
We preprocessed the data using Independent Component Analysis to re- move mostly eye movement artifacts. Then, we extract several features mostly related to the frequency domain of the signal. Finally usingWaikato Environment for Knowledge Analysis (WEKA), we tried several classifiers method to know which one give a better result.
In our experiment, we conducted three experiments in which three stress level were compared; no-stress, average and high-stress level. We were able to classify player state with an average of 79.089 % in accuracy level using Decision Tree classifier. We also performed a comparison between classifying 3 user state and with pair-wise classification (only two states). On average, we achieved 78.7864 % for distinguishing three classes of states. While, classifying two-states achieved an average of over than 80 % in accuracy level.