Deep ConvLSTM With Self-Attention for Human Activity Decoding Using Wearable Sensors

Decoding human activity accurately from wearable sensors can aid in related to healthcare and context awareness. The present in this domain use recurrent and/or convolutional models to capture spatio-temporal features from time-series data from multiple sensors. We a deep neural network architecture that not only captures the-temporal features of multiple sensor time-series data but also selects, important time points by utilizing a self-attention mechanism. We show validity of the proposed approach across different data sampling strategies six public datasets and demonstrate that the self-attention mechanism gave a improvement in performance over deep networks using a combination recurrent and convolution networks. We also show that the proposed approach a statistically significant performance enhancement over previous-of-the-art methods for the tested datasets. The proposed methods open for better decoding of human activity from multiple body sensors over periods of time. The code implementation for the proposed model is at https://github.com/isukrit/encodingHumanActivity.