Learning and Annotating Activities for Home Automation using LSTM

Driven by the growth in ubiquitous sensor networks and IoT, Human Activity Recognition (HAR) has emerged to be an exciting challenge. Popular machine learning approaches proposed in this direction have shown promise with modest performance. Treatment of HAR data in a recurrent form, and subsequent analysis using deep networks such as RNNs/LSTM is largely unexplored. In this work, we treat HAR data stream, generated from a smart home setup, as a series of events and apply LSTM to perform feature extraction. We use the Aruba dataset from the CASAS project [4], where we apply LSTM to extract features and perform annotations. We then apply standard classification techniques to recognize the different activities in the newly annotated data. We observe that, using our method we can achieve annotation accuracies of up to 79.5% which is 13.6% better than other state-of-the-art solutions.