Recognition of Human Activities in Smart Homes Using Stacked Autoencoders

There is a growing interest in the domain of smart homes. One of the most important tasks in this domain is the recognition of inhabitants' activities. To ameliorate the proposed approaches, we propose, in this paper, a Staked Autoencoder (SAE) algorithm based on a deep learning framework for recognizing activities in a smart home. Our approach is tested on the Washington State University (WSU) dataset. We will show that our proposed approach outperforms existing methods such as the Artificial Neural Networks (ANNs) in terms of recognition accuracy of activities. In particular, the SAE shows an accuracy of 87.5% in recognizing activities based on WSU smart home dataset while the ANN algorithm has shown an accuracy of 79.5% on the same dataset.