CONTEXT-AWARE HEALTH MONITORING USING FEDERATED LEARNING

Abstract

Recent advances in the Internet of Things (IoT) and communication technologies have enabled the emergence of the Internet of Medical Things (IoMT), which integrates IoT into healthcare for real-time monitoring and intelligent intervention. IoMT applications are increasingly used for continuous health tracking, but traditional machine learning (ML) approaches—dependent on centralized data collection—raise critical concerns around patient privacy, data security, and regulatory compliance. Federated Learning (FL) addresses these issues by enabling collaborative model training across stakeholders (e.g., hospitals, homes) without sharing sensitive raw data, thereby preserving privacy and supporting distributed intelligence within IoMT ecosystems. This project presents a privacy-preserving health monitoring and emergency detection system that integrates IoT devices, FL, and Active Learning (AL). The system uses a two-stage architecture: physiological and contextual data (heart rate, SpO2, ECG, age, gender) are collected via wearable sensors, while multimedia devices perform Human Activity Recognition (HAR) for anomaly validation. FL ensures secure model training across distributed sources. Experimental results show an anomaly detection accuracy of ˜0.89 using SensorECGCNN Light and ˜0.85 HAR accuracy using a Temporal Convolutional Network (TCN). A Meta Federated Model (MFL) combines these outputs, while AL enables model refinement through human feedback, improving adaptability and system robustness over time.