Continuous Wearable Signal Processing and Multi-Aspect User Feedback Modeling via Distributed Edge–Cloud Transformer-LSTM Architectures

• Billy Elly

  Lautech

  Author

The proliferation of wearable biosensors has enabled continuous physiological monitoring, yet existing systems face fundamental limitations in balancing real-time processing latency, long-term temporal pattern recognition, and personalized user feedback integration. Current approaches typically sacrifice either computational efficiency for deep contextual modeling or personalization for generalized population-level analytics. This study addresses these gaps by proposing a distributed edge–cloud architecture that synergistically integrates Transformer and Long Short-Term Memory (LSTM) networks for hierarchical wearable signal processing. The framework deploys lightweight LSTM models on edge devices for real-time anomaly detection (achieving 48 ms inference latency), while cloud-based Transformer models perform deep contextual analysis of longitudinal health trajectories (95.3% classification accuracy for adverse cardiac events, p < 0.001). A multi-aspect user feedback loop incorporates physiological, behavioral, and subjective well-being signals through a cross-modal fusion mechanism, enabling personalized adaptive calibration. The hybrid architecture achieves 89.4% improvement in early warning lead time over baseline single-model approaches and demonstrates 97.2% accuracy in personalized health state prediction across 1,200 subject-months of wearable data. This framework establishes a replicable paradigm for continuous health monitoring that balances computational efficiency with contextual depth, with significant implications for remote patient monitoring, chronic disease management, and proactive intervention systems

• pdf

Copyright (c) 2026 Billy Elly (Author)

This work is licensed under a Creative Commons Attribution 4.0 International License.