Random forests based recognition of human activities and postural transitions on smartphone

Postural transitions are transitory movements with limited duration bounded by two physical activities, which represent the transition between two activities. Recognition of postural transitions is a challenging task due to low occurrence rate of transitions and short duration than physical activities. Postural transitions may decrease recognition accuracy when postural transitions are unspecified in the recognition system. This paper deals with the occurrence of postural transitions during activity performance. Hence, this paper proposes importance score driven random forests to recognize activities and postural transitions on the smartphone. The proposed model first selects relevant features rely on the relative importance of features, and then feed selected features to random forests. This model shows 100% recognition accuracy on a benchmark dataset and indicates that features created from accelerometer signals are more relevant for recognition than features created from gyroscope signals.