🤖 AI Summary
To address the challenges of continuous physical activity monitoring and early detection of behavioral changes in multimodal IoT systems, this paper proposes an information fusion theory and an unsupervised behavioral clustering method integrating heterogeneous multimodal sensor data. Our approach uniquely combines temporal feature extraction, a hybrid K-means–DBSCAN clustering algorithm, and an edge–cloud collaborative computing architecture to enable fine-grained, adaptive modeling of daily activity patterns. Evaluated in real-world in-home settings with older adults, it achieves 92.3% accuracy in behavioral stage identification and reduces average behavioral anomaly detection latency to 1.8 hours. The method significantly enhances behavioral similarity mining, interpretable feature extraction, and early intervention capability for chronic disease risk. By unifying sensing, learning, and distributed computation, it establishes a scalable technical paradigm for longitudinal health monitoring in smart living environments.
📝 Abstract
This study exploits information fusion in IoT systems and uses a clustering method to identify similarities in behaviours and key characteristics within each cluster. This approach facilitates early detection of behaviour changes and provides a more in-depth understanding of behaviour routines for continuous health monitoring.