Higher-intensity activity and greater movement distance were the strongest drivers of calorie expenditure in a machine learning (ML) analysis of wearable device data, with support vector regression outperforming other models in predicting energy use.
In a study published in Scientific Reports, investigators from the Institute of Physical Education Teaching and Research at Fuzhou University in Fujian, China, analyzed physical activity data from 30 patients using wearable fitness trackers to evaluate how different activity metrics contribute to calorie consumption. The findings showed that total distance and total steps were the most influential predictors, while time spent in high-intensity activity further amplified energy expenditure.
The study used a publicly available dataset collected over a two-month period, including daily step count, activity intensity, and calorie expenditure. After preprocessing, researchers selected features strongly associated with calorie expenditure, including total steps, total distance, and activity intensity measures, and split the data into training and testing sets in a 7:3 ratio. Four machine learning models were compared: support vector regression, radial basis function neural network, random forest, and extreme gradient boosting.
Support vector regression performed best overall. On the test set, it had an R² of 0.78, compared with 0.75 for the neural network, 0.67 for random forest, and 0.63 for gradient boosting. It also had lower prediction errors, with a root mean square error of 329 and a mean absolute error of 230. Although gradient boosting performed well during training (R² of 0.94), its lower test performance suggests it may have been overfitting.
A SHAP (Shapley Additive Explanations) analysis showed that total distance had the biggest impact, followed by total steps. Higher values for both variables were associated with increased predicted calorie expenditure. High-intensity activity such as “very active minutes” and “fairly active minutes” also played an important role, while low-intensity activity and sedentary time had little impact.
The analysis showed that high-intensity activity was more strongly linked to calorie burn than moderate activity. More very active minutes led to larger increases in calories burned, while moderate activity had a smaller but still positive effect. Light activity demonstrated only a limited association, and sedentary behavior showed little to no consistent relationship with calorie expenditure.
The relationship between activity and calories burned is not always straightforward, according to the analysis. Step count alone did not consistently increase calorie burn; it depended on total distance. When step counts were higher without a matching increase in distance, the predicted increase in calories burned was smaller. This suggests that how efficiently and intensely someone moves affects how their steps translate into energy use.
Subgroup analysis showed that high-intensity activity had a stronger effect on energy use than other activity levels. When activity went above average levels, calorie burn increased more sharply, suggesting a nonlinear response. Moderate activity had a smaller impact, especially in people already doing high-intensity activity. Sedentary time had little direct effect but may still influence energy use when combined with other activity patterns.
Researchers noted several limitations, including a small sample size and missing data on physiological factors like heart rate and age. The analysis also relied on a single dataset without external validation, which may have limited generalizability.
The findings suggest that both activity volume and intensity are central to energy expenditure, with implications for personalized exercise planning. “High-intensity activities and greater activity distances should be focal points in health management and exercise interventions,” the authors concluded.
Study authors reported no external funding and no conflicts of interest.
