Accuracy of Samsung Gear S Smartwatch for Activity Recognition: Validation Study

Area of Science:

• Biomedical Engineering
• Wearable Technology
• Physical Activity Research

Background:

• Wearable accelerometers are crucial for physical activity measurement.
• Smartwatches offer potential for research due to built-in accelerometers, but require validation.

Purpose of the Study:

• To validate Samsung Gear S (SGS) smartwatch accelerometer data against a research-grade ActiGraph GT3X+ (GT3X+).
• To compare SGS and GT3X+ in activity recognition, movement detection, intensity, locomotion, and METs estimation.

Main Methods:

• Simultaneous data collection from SGS and GT3X+ on a mechanical shaker, treadmill, and during 15 daily activities.
• Feature extraction and machine learning models (random forest, SVM, neural networks, decision trees) for 6 classification tasks.
• Evaluation of feature extraction window lengths and METs estimation accuracy.

Main Results:

• High correlation (≥.89) between SGS and GT3X+ on shaker/treadmill tests; ≥.70 for daily activities (except computer work).
• Classification accuracies: activity intensity (0.87), locomotion (1.00), sedentary (0.98), movement location (0.85), activity recognition (0.64).
• No significant differences between SGS and GT3X+; Random Forest optimal for METs estimation.

Conclusions:

• Commercial smartwatches, such as the Samsung Gear S, are valid alternatives to research-grade monitors.
• SGS smartwatches can reliably be used for activity recognition, movement detection, intensity, and locomotion tasks in research settings.