PPG Quality Segmentation - Signals and Labels

Two publicly available datasets, the PASS and EmpaticaE4Stress databases, were utilised in this study. They were chosen because they both used the same Empatica E4 device, which allowed the acquisition of a variety of signals, including PPG and EDA. The dataset consists of in 1587 30-second PPG segments. Each segment has been filtered and normalized using a 0.9–5 Hz band-pass and min-max normalization scheme. The dataset contains the original data, the motion artifacts segmentation labels, binary class labels, source dataset name, subject, % of corrupted signals, and the split class (train, validation, and test). Furthermore, there are two other small datasets, one is a subset of the original one with human labels (from both expert and non-expert annotators) and the other contains data from two open-source devices (Bangle.js 2 and EmotiBit) collected in real life scenarios.