Age Differences in Estimating Physical Activity by Wrist Accelerometry Using Machine Learning.

Area of Science:

• Biomedical Engineering
• Sports Science
• Machine Learning

Background:

• Wearable fitness trackers and smartwatches are increasingly popular for health monitoring.
• Accurate measurement of physical activity across all adult age groups remains a challenge.

Purpose of the Study:

• To evaluate machine learning models for recognizing physical activity types, intensities, and estimating energy expenditure.
• To assess model performance across young, middle-aged, and older adults using accelerometer data.

Main Methods:

• Four machine learning models (decision tree, random forest, XGBoost, LASSO) were trained.
• Models used wrist-worn tri-axial accelerometer data from 253 participants (aged 20-89).
• Performance was evaluated for activity type, intensity, energy expenditure, and individual activity recognition.

Main Results:

• Machine learning models demonstrated high accuracy in recognizing physical activity type and intensity, and estimating energy expenditure.
• XGBoost models achieved high F1-Scores for sedentary, locomotion, and lifestyle activity types.
• Performance showed minimal differences across young, middle-aged, and older adult groups.

Conclusions:

• Machine learning models are accurate for labeling physical activity types, intensity, and energy expenditure.
• These models show consistent performance across different adult age groups.
• Further research may be needed to optimize individual physical activity recognition.