Back

Development of machine learning-based algorithms for classifying physical activity intensity using wrist and thigh worn wearables

Koemel, N. A.; Biswas, R. K.; Ahmadi, M.; Stamatakis, E.

2024-08-09 public and global health
10.1101/2024.08.07.24311585 medRxiv
Show abstract

The use of semi-supervised learning approaches can be used to extend a base-level classifier and offers a significant advantage by reducing the need for extensive labeled datasets. We utilized a two-stage semi-supervised learning model to classify physical activity intensity for wrist and thigh worn monitors, by retraining a base classifier with free-living wearable sensor data. Data was collected in two-phases comprising a laboratory and free-living session. Total classified time spent in light intensity, moderate intensity, and vigorous intensity were not significantly different from ground-truth minutes for either placement. The machine learning classifiers re-trained on free-living data accurately predicted light, moderate and vigorous intensity between both device placements. These findings demonstrate that similar estimates of physical activity intensity can be correctly classified for wrist and thigh placements when using semi-supervised techniques.

Matching journals

The top 5 journals account for 50% of the predicted probability mass.

50% of probability mass above

"Similar papers" are the closest papers from that journal in the model's embedding space. They show what the match is built on, but the ranking comes mostly from a classifier over the whole training set, not from these examples alone.