Ensemble Feature Selection for the Recognition of Human Activities and Postural Transitions on Smartphones

Human activity recognition (HAR) plays a vital role in the field of ambient assisted living for the welfare of senior citizens which provides satisfactory results in terms of wearable devices, ambient sensors, and smartphones. The usage of smartphones is familiar among users and it is widely adopted in activity recognition due to its low-cost, intrusiveness, and convenience. The embedded inertial sensors such as tri-axial accelerometer and gyroscopes are utilized to analyze the activities through the generated time and frequency domain features. Various state-of-the-art techniques have been proposed for significant feature selection to improve recognition performances. This paper proposes an ensemble of feature selection technique which selects minimal features from the intersection of features evolved in various other feature selection techniques. Extensive experimentation has been carried out using the HAPT dataset to identify the significant difference with state-of-the-art techniques through familiar classification performance metrics.