Designing Robust Machine Learning Model for Enhanced Human Activity Recognition
Pradeep Kumar Sharma, Harsh Mathur · Communications on Applied Electronics · 2025
For health monitoring and fitness tracking, wearable computing, smart settings, and healthcare require effective human activity recognition (HAR).HAR systems must work in real time despite noisy sensor input, changing surroundings, and other obstacles.The Spatio-Temporal Attention-based Hybrid Neural Network (STAHNN) improves HAR task robustness and efficiency.STAHNN uses CNNs to extract spatial characteristics and RNNs to model temporal dependencies.Self-attention reduces noise by focusing on relevant spatiotemporal elements.Jittering and scaling promote generalization, whereas domain adaptation procedures reduce sensor variability and ensure performance.STAHNN outperforms other HAR techniques in noisy data management and activity pattern adaptation, according to extensive studies.STAHNN solves HAR problems scalable and robustly by bridging laboratory and real-world performance.