Consideration of human motion’s individual differences-based feature space evaluation function for anomaly detection

There are many researches of various human activity recognitions from the data of inertial sensors by using machine learning. In these researches, it is important to consider the individual difference. Even if the subjects perform the same activity, the data obtained from each subject are of different behaviors. Thus, if we construct the feature space, there is a possibility that the human activity of each subject does not concentrate on one region. In this case, if we consider the anomaly detection of the human activities by using this space, it is difficult to draw the boundary between the normal and anomaly activities. Therefore, the evaluation index/function that can search better feature values/space for various people is necessary. In this paper, we propose an evaluation function named “Consideration of Human motion’s Individual differencesbased Feature Space (CHI-FS) evaluation function” for the anomaly detection. We also confirm the effectiveness of the CHI-FS evaluation function by using the simulation data and the data of inertial sensors during car driving.