Fig. 5: Artefact-free fatigue profiling in daily life.
a, Photograph of a participant wearing the MAP for real-time physiological signals monitoring. The mobile phone shows the user interface. Scale bar, 10 cm. b, Robustness evaluation of collected different physiological signals from MAP and a commercial sensor. Data were obtained from independent participants (n = 5). Bars indicate mean ± s.d. c, Frequency histogram of the difference of adjacent RR peaks (Ri+1–Ri) at normal and fatigue states. The narrower distribution indicates the reduced HRV. d, Continuous 3-day multimodal physiological signals monitoring during different daily activities on a participant. For each timepoint, and three segments within this window (over 2 min) were analysed. Error bars represent the mean ± s.d. calculated from these replicates (n = 3). e, SHAP summary plot for fatigue evaluation based on the dataset collected by the MAP. Each axis plots the distribution of SHAP value of the selected feature for predicted instance. f, Ture value versus assessed fatigue levels. Data are presented with a ±2-point buffer to account for potential errors in the FAS questionnaires. g, ROC curves evaluating the classification performance of the MAP system for fatigue state profiling. The curves are colour-coded from light to dark to represent increasing score scales on the fatigue assessment. h, t-SNE-based clustering of fatigue levels among different participants. A total of five participants were assessed, and their fatigue states were categorized into seven levels based on scores from the FAS. Activity icons in d created in BioRender; Tian, G. https://biorender.com/omp1s6v (2026).
