Improving Deep Learning for HAR with shallow LSTMs

With this paper, we aim at challenging this belief and suggest in HAR is the DeepConvLSTM. In this paper we propose to alter that re-examining the architecture of the DeepConvLSTM by the DeepConvLSTM architecture to employ a 1-layered instead employing a one-layered LSTM, has considerable benefits. of a 2-layered LSTM. We validate our architecture change on 5 Our paper's contributions are threefold: publicly available HAR datasets by comparing the predictive performance with and without the change employing varying hidden (1) We show that an altered DeepConvLSTM architecture with units within the LSTM layer(s). Results show that across all datasets, a one-layered LSTM overall outperforms architectures employing our architecture consistently improves on the original one: Recognition a two-layered LSTM, by validating our claim using performance increases up to 11.7% for the F1-score, and our the Opportunity dataset [29] as seen in [26], as well as 4 architecture significantly decreases the amount of learnable parameters.