Human Activity Recognition and Feature Selection for Stroke Early Diagnosis

Human Activity Recognition (HAR) refers to the techniques for detecting what a subject is currently doing. A wide variety of techniques have been designed and applied in ambient intelligence -related with comfort issues in home automation- and in Ambient Assisted Living (AAL) - related with the health care of elderly people. In this study, we focus on the diagnosing of an illness that requires estimating the activity of the subject. In a previous study, we adapted a well-known HAR technique to use accelerometers in the dominant wrist. This study goes one step further, firstly analyzing the different variables that have been reported in HAR, then evaluating those of higher relevance and finally performing a wrapper feature selection method. The main contribution of this study is the best adaptation of the chosen technique for estimating the current activity of the individual. The obtained results are expected to be included in a specific device for early stroke diagnosing.