Motion recognition provides movement information for people with physical dysfunction, the elderly and motion-sensing games production, and is important for accurate recognition of human motion. We employed three classical machine learning algorithms and three deep learning algorithm models for motion recognition, namely Random Forests (RF), K-Nearest Neighbors (KNN) and Decision Tree (DT) and Dynamic Neural Network (DNN), Convolutional Neural Network (CNN) and Recurrent Neural Network (RNN). Compared with the Inertial Measurement Unit (IMU) worn on seven parts of body. Overall, the difference in performance among the three classical machine learning algorithms in this study was insignificant. The RF algorithm model performed best, having achieved a recognition rate of 96.67%, followed by the KNN algorithm model with an optimal recognition rate of 95.31% and the DT algorithm with an optimal recognition rate of 94.85%. The performance difference among deep learning algorithm models was significant. The DNN algorithm model performed best, having achieved a recognition rate of 97.71%. Our study validated the feasibility of using multidimensional data for motion recognition and demonstrated that the optimal wearing part for distinguishing daily activities based on multidimensional sensing data was the waist. In terms of algorithms, deep learning algorithms based on multi-dimensional sensors performed better, and tree-structured models still have better performance in traditional machine learning algorithms. The results indicated that IMU combined with deep learning algorithms can effectively recognize actions and provided a promising basis for a wider range of applications in the field of motion recognition.

Dynamic oxygen uptake (VO₂) reflects moment-to-moment changes in oxygen consumption during exercise and underpins training design, performance enhancement, and clinical decision-making. We tackled two key obstacles—the limited fusion of heterogeneous sensor data and inadequate modeling of long-range temporal patterns—by integrating wearable accelerometer and heart-rate streams with a convolutional neural network–LSTM (CNN-LSTM) architecture and optional attention modules. Physiological signals and VO₂ were recorded from 21 adults through resting assessment and cardiopulmonary exercise testing. The results showed that pairing accelerometer with heart-rate inputs improves prediction compared with considering the heart rate alone. The baseline CNN-LSTM reached R² = 0.946, outperforming a plain LSTM (R² = 0.926) thanks to stronger local spatio-temporal feature extraction. Introducing a spatial attention mechanism raised accuracy further (R² = 0.962), whereas temporal attention reduced it (R² = 0.930), indicating that attention success depends on how well the attended features align with exercise dynamics. Stacking both attentions (spatio-temporal) yielded R² = 0.960, slightly below the value for spatial attention alone, implying that added complexity does not guarantee better performance. Across all models, prediction errors grew during high-intensity bouts, highlighting a bottleneck in capturing non-linear physiological responses under heavy load. These findings inform architecture selection for wearable metabolic monitoring and clarify when attention mechanisms add value.

Background: In the fields of rehabilitation, public health, military training and other domains, the accurate and effective monitoring of exercise intensity during exercise can control the occurrence of sports injuries, which is of great significance for people’s healthy lives. Objective: This study combined easily collectable multi-dimensional sensor data and various algorithm models to achieve cross-individual recognition of low, middle and high levels of exercise intensity. Methods: This study compared the recognition performance of different algorithm models using acceleration and angular velocity sensors worn on seven body parts through individualised body data characteristics. Results: The recognition performances of two classical machine learning algorithms were the worst, with a recognition rate of only 82.97% and 80.31%. The performances of two ensemble learning algorithms were slightly better, with a recognition rate of 88.86% and 87.35%. The deep sub-domain adaptation network algorithm proposed in this study exhibited the best performance, with a recognition rate of 92.87%. This study utilised multi-dimensional sensors to construct a cross-individual exercise intensity recognition model for different parts of the body, and the overall recognition rate of the left part was higher than that of the right part. Moreover, the recognition effect upon wearing sensors on the left side of the body is better than the right in running events. Conclusions: The results of this study have demonstrated the effectiveness of combining domain adaptation methods and multi-dimensional sensors for cross-individual exercise intensity recognition, laying a solid theoretical foundation for broader-scale cross-individual exercise intensity recognition in future research.

The drop landing (DL) process is a complex neuromuscular control process influenced by multiple factors, among which fatigue state and landing height play significant roles. This study aimed to investigate the influence mechanisms of these two factors on lower-limb landing strategies and to examine the feasibility of explainable machine learning methods in analyzing landing strategies. Joint angles and joint moments of the hip, knee, and ankle in the sagittal plane, as well as vertical ground reaction force (vGRF), were collected under different landing conditions (40 cm and 80 cm heights and their corresponding fatigue states). Cohen’s d effect size analysis was applied to examine group differences in each feature across the time series, and seven classification models were constructed based on multi-dimensional biomechanical features to identify different landing states. Meanwhile, SHapley Additive exPlanations (SHAP) method was employed to analyze feature importance and interactions. Results showed that the Light Gradient Boosting Machine achieved classification accuracies above 0.915 across all four tasks, verifying the feasibility of machine learning methods for landing strategy analysis. Temporal effect size analysis revealed that differences induced by changes in landing height were mainly concentrated in the early landing phase (0%–40%). In contrast, the fatigue effect resulted in a wider distribution and higher magnitude of regions with large effect sizes. SHAP analysis indicated that the knee joint moment had the highest importance across all classification tasks, vGRF contributed prominently in fatigue-related tasks, and feature interaction patterns differed significantly across tasks. In conclusion, explainable machine learning methods can effectively identify and explain differences in DL strategies under various landing conditions, providing a theoretical basis and methodological support for sports injury risk assessment and training strategy optimization.