Workplace activity classification from shoe-based movement sensors.

Area of Science:

• Occupational health
• Wearable technology
• Machine learning for activity recognition

Background:

• High occupational physical activity is linked to adverse health outcomes.
• Shoe-based sensors offer objective physical activity measurement but require validation in free-living settings.
• Investigating the feasibility and accuracy of shoe sensor-based activity classification in industrial work environments.

Purpose of the Study:

• To assess the feasibility of using shoe-based sensors for activity classification in an industrial setting.
• To evaluate the accuracy of machine learning models for classifying occupational physical activity.
• To compare sensor-based classification performance in lab versus free-living conditions.

Main Methods:

• Trained three machine learning models (random forest, support vector machine, k-nearest neighbour) using lab-calibrated data.
• Validated models with 29 industry workers, comparing sensor data to observer notes.
• Focused on classifying activities like walking and stationary postures.

Main Results:

• The random forest classifier showed superior performance in free-living validation.
• Initial random forest accuracy was 83% (lab) and 43% (free-living).
• Combined activities improved accuracy to 96% (lab) and 71% (free-living); 99% of samples were stationary or walking.

Conclusions:

• Shoe-based movement sensors can accurately classify walking and stationary activities in occupational settings.
• Activity distribution in the workplace is crucial for validating classification models.
• This technology holds promise for objective assessment of physical activity at work.