Researchers have developed ActiNet, an open-source tool that utilizes self-supervised deep learning for classifying activity intensity from wrist-worn accelerometry data. The ActiNet model, comprising HARNet and hidden Markov model smoothing, achieved a mean macro F1 score of 0.82 and a Cohen's kappa score of 0.86. This performance surpasses the baseline random forest plus HMM approach, demonstrating consistent improvements across different age and sex subgroups. The findings suggest ActiNet's utility for future epidemiological studies analyzing physical activity and health. AI
Summary written by gemini-2.5-flash-lite from 1 source. How we write summaries →
IMPACT Provides a new open-source tool for epidemiological studies to extract activity intensity labels from wearable sensor data.
RANK_REASON This is a research paper detailing a new open-source tool for activity recognition using deep learning.