Sign in Subscribe
Exercise

Machine Learning Model for Physical Activity Classification

Strategic Insights

— 1 min read

Researchers from Chongqing Normal University have developed a model for classifying physical activity types using healthcare wearable devices and optimized machine learning (ML) methods. The model, which utilizes XGBoost as the primary ML algorithm, has potential applications in continuous activity monitoring for rehabilitation and remote patient care in clinical settings. The study aimed to optimize the model's performance by fine-tuning its hyperparameters using six meta-heuristic algorithms. The results showed a significant improvement in classification performance, surpassing other models in all evaluation metrics.

Key Takeaways:

  • The study aimed to develop a model for classifying physical activity types using healthcare wearable devices and optimized ML methods.
  • The developed model has potential applications in continuous activity monitoring for rehabilitation and remote patient care in clinical settings.
  • The dataset comprised sensor data collected from multiple sessions with annotated samples across six activity classes: Lying, Self Pace Walk, Sitting, Running 3 METs, Running 5 METs, and Running 7 METs.
  • Four primary ML algorithms were used: HistgradientBoosting, Random Forest Classifier, Extreme Gradient Boosting (XGBoost) Classifier, and Decision Tree.
  • Evaluation results showed that XGBoost performed the best in terms of accuracy, with an 0.915644 accuracy rate.
  • The study found that the SADE + XGBoost hybrid model outperformed the others, achieving the highest accuracy.
  • The proposed optimization resulted in approximately 1% accuracy improvement over the baseline XGBoost model.
  • The optimized model's results demonstrated a significant improvement in classification performance, surpassing other models in all evaluation metrics.

Statistics:

  • Accuracy: 0.915644
  • Recall: 0.916451
  • Precision: 0.914528
  • F1 Score: 0.91534
  • Percentage of accuracy improvement over baseline XGBoost model: approximately 1%

Sources:

  • "Prediction of Physical Activity Types Based On Healthcare Wearable Devices Using Optimized Machine Learning Methods" (Signal, Image and Video Processing, 2025;19(13))
  • Wei Si, Chongqing Normal University (Chongqing 130117, People's Republic of China)

Read more