As an indicator regarding cardiovascular health, vascular dynamics, and more, Photoplethysmography (PPG) consists of a baseline with two pulsatile peaks that reflect the blood flow variations. In the absence of critical studies and a precise understanding of PPG signal this study proposes a deep learning approach to predict and define each phase of PPG waveform utilizing Gated recurrent unit (GRU) due to its optimal performance in the short-term dependencies in sequences, the model is composed mainly of GRU layer with 20 hidden units applying tanh and sigmoid activation function, followed by dropout layer and a fully connected layer, afterward a softmax layer was added, we conducted our study on a dataset combined of an accessible Bed-Based Ballistocardiography Dataset (BBB), after that, the data was manually labeled. Following up, hyperparameters such as learning rate, batch size, and other parameters were adjusted after numerous trials, which resulted in test accuracy, precision, and recall of 0.875, 0.892, and 0.854, respectively in the used dataset. These promising results might be the first step of more future investigations concerning PPG.
