Influencing Factors Mining and Modeling of Energy Expenditure in Running Based on Wearable Sensors

Real-time monitoring of energy expenditure plays a critical role in enhancing both athletic performance and personal fitness management in running sports. It provides runners with immediate insights into their energy expenditure, allowing them to optimize their pace, intensity, and technique during training or competition. With the development of wearable technology, wearable devices have been widely used in sports energy expenditure assessment. However, most of the existing wearable energy expenditure methods rely on deep learning, which tends to have poor interpretability in terms of model explainability. In this study, we analyze features related to energy expenditure in running sports from multiple dimensions, including demographics, physical activity, and physiological metrics. Based on the highly interpretable regression algorithms, we develop an accurate energy expenditure computing model. We propose a hand-crafted feature selection method and selected 743 features for use in the model. Using the selected features, we build energy expenditure computing models with several machine learning algorithms, including linear regression (LR), decision tree (DT), K-nearest neighbor (KNN), random forest (RF), support vector regression (SVR) and gradient-boosted tree (GBR). Among these, the GBR achieves the best performance, with a correlation coefficient (CC) of 0.970, a root mean square error (RMSE) of 1.004, and a mean absolute error (MAE) of 0.729, tested by five-fold cross-validation on a dataset with 34 volunteers. These results indicate that the features extracted in this study are significant for real-time and accurate prediction of running energy expenditure.