Human Activity Prediction-aware Sensor Cycling in Smart Home Networks

Activity recognition in smart homes required extensive machine learning methods operating over data obtained from the sensors connected to various things at homes. However, human behavior is always dynamic and there is a possibility of missing important activities. Similarly, the sensors operate over battery power, and detecting frequent activities can drain the battery eventually. In this paper, we propose Human Activity Predication-aware Sensor Cycling (HAPSC) approach that efficiently detects the future activities and designates the sensors to each predicated activity using a hybrid Long-Short Term Memory (LSTM) model with the Quality Learning (QL) support. The QL is added to detect those activities which are missed during the training process. The simulation is performed in comparison with the leading machine learning algorithms for the detection accuracy and energy consumption of the sensors. The performance evaluation shows that the proposed scheme detects the activities with an accuracy of 85.7% to 99.4%. Similarly, the energy consumption of the sensors and the network lifetime is significantly improved compared to existing duty cycling schemes.