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

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