An Enabling Technology for Industry 5.0: Human Activity Recognition Based on Differential Sampling and Hybrid Parameter Seeking

ABSTRACT Within the framework of the Industrial Internet of Things (IIoT) and the human‐centric vision of Industry 5.0, the integration of intelligence into human activity recognition (HAR) represents a prevailing trend. To address this objective, an advanced machine learning classification approach is proposed. Primarily, targeting challenges induced by imbalanced sample distributions, a sliding window mechanism was implemented to extract activities characterized by larger sample sizes. Subsequent feature engineering involved applying weighted summation and differentiation operations to highly correlated features. To ensure equitable classification performance across categories, dual allocation of training weights per class was executed. Finally, a hybrid search strategy determined optimal hyperparameters for the eXtreme Gradient Boosting (XGBoost) model. Validation across datasets from IIoT‐enabled wearable sensors demonstrated peak performance metrics: Precision reached 0.881, Recall attained 0.882, F 1‐Score achieved 0.881, and Accuracy hit 0.882. Analysis revealed that weighted ensemble of feature attributes with strong correlation helped improve the model's ability to recognize categories. This method lays the foundational technical framework for AI‐driven HAR within the industrial IoT ecosystem.