A new preprint on arXiv presents a group-aware comparative study of classical machine learning and deep learning methods for sensor-based human activity recognition (HAR). The work focuses on wearable or smartphone inertial sensors, a common setting for HAR research.
The abstract notes that reported performance in this area is often evaluated with both classical and deep learning approaches, but that results can be strongly affected by sensor-related factors. The full details of which factors and how they are handled are not visible in the truncated abstract, so the specific methodology and findings remain unclear from the available text.
The 'group-aware' framing suggests the authors account for variations across groups or conditions, but the source does not specify what those groups are. Readers interested in the complete analysis will need to consult the full paper on arXiv.