Deep Learning Model for ECG Analysis and Prediction Using mHealth Datasets

The popularity of Human Activity Recognition (HAR) as a research topic has increased because of its many applications. New approaches to HAR concerns have emerged as a result of deep learning. Here, a deep network architecture based on residual bidirectional Long Short-Term Memory (LSTM) is recommended. The ability of a bidirectional link to integrate the forward state of positive time and the backward state of negative time is one advantage of the new network. Second, the gradient vanishing problem is successfully solved because remaining connections between stacked cells allow gradients to function as a shortcut. In order to raise the recognition rate, the proposed network generally shows improvements on the spatial (stacked residual connections) and theoretical (using bidirectional cells) dimensions. Testing with the opportunity dataset and the public domain UCI dataset greatly increases accuracy in comparison to earlier findings.