4.1.4. Accelerometer Feature Extraction Methods

Research has shown that movements of the human body and postures can indeed be employed as a means to detect signs of different emotional states. The dynamics of body movement were investigated by Castellano et al. who used multimodal data to identify human affective behaviors. Specific movement metrics, such as the amount of movement, intensity and fluidity, were used to help deduct emotions, and it was found that the amount of movement was a major factor in distinguishing different types of emotions [51]. Melzer et al. investigated whether movements comprised of collections of Laban movement components could be recognized as expressing basic emotions [52]. The results of their study confirm that, even when the subject has no intention of expressing emotions, particular movements can assist in the perception of bodily expressions of emotions. Accelerometer sensors may be used to detect these movements and different types of affect. The accelerometer sensor data are used for two different purposes in our system. Firstly, we extracted features from the accelerometer sensor, for detecting stress levels. We also selected the features to be used as described in Table 4 [53] and, as mentioned above, this sensor was also employed to clean the EDA signal in the EDAExplorer Tool [41].

**Table 4.** ACC features and their definitions.

