Physiological signals based fatigue prediction model for motion sensing games

We present a fatigue prediction model for motion sensing games, dependent on the change of physiological signals including blood volume pulse, skin conductance, respiration, skin temperature and electromyography (EMG). After extracting a range of features followed by using sequential floating forward selection (SFFS) to select features, support vector regression (SVR) was used to construct our prediction model that can predict how long participants enter fatigue states. The root mean square error (RMSE) and the relative root square error (RRSE) of our model are respectively 198.36s and 0.51 for subject-dependent, and 522.94s and 0.97 for subject-independent. The results indicate each subject has individualized physiological pattern when they felt fatigue.