Informative sensor selection on clustered sensors
Saed Sa’deh Juboor, Sook-Ling Chua, Lee Kien Foo · Journal of Physics Conference Series · 2019
Many researchers have focused their work on recognising activities in smart homes, with the aim to support the occupants by monitoring their daily activities and identify any abnormalities. In order to recognise human daily activities, sensors are installed in the home to collect information about the occupant. Activities can then be inferred from a sequence of sensor observations output from the house. However, one challenge still remains: which sensors are useful to effectively recognise the occupant's activities. Many traditional filter-based methods have been proposed in the literature for sensor selection but these methods may not consider sensor inter-redundancy. Motivated by this, this paper addresses the sensor selection problem using clustering and then apply the filter-based method on the clustered sensors. The effectiveness of the method is evaluated using two well-known public smart home datasets. The results showed that our proposed method not only able to reduce the number of sensors needed but also able identify the set of sensors that are useful for activity recognition.