IoT-Based Elderly Health Monitoring Information System with Federated Learning for Early Health Risk Prediction

Abstract

The growing elderly population in Indonesia is directly associated with an increasing burden of non- communicable diseases such as hypertension, heart disease, diabetes mellitus, and chronic obstructive pulmonary disease (COPD), creating a need for continuous rather than periodic health-monitoring mechanisms. This study aims to design an Internet of Things (IoT)-based elderly health-monitoring information system integrated with a Federated Learning (FL) approach to perform early prediction of health risk without compromising the privacy of users' medical data. The research method used is Design Science Research (DSR), comprising problem identification, requirement elicitation, design, implementation, and evaluation stages. A hardware prototype was built using an ESP32 microcontroller connected to a MAX30102 sensor (heart rate and oxygen saturation), a blood pressure sensor, a DS18B20 body-temperature sensor, and an MPU6050 sensor for fall detection, which transmit data in real time to a web-based dashboard via the MQTT protocol. The health-risk prediction process is trained in a distributed manner on user devices using the Federated Averaging (FedAvg) scheme, so that raw data never leaves the device. Preliminary testing results show that the system is able to display vital-sign data in real time on the dashboard and classify risk status (normal, alert, at-risk). The application of Federated Learning is shown to improve privacy protection of elderly patients' medical data without significantly reducing classification-model performance, so that the system has the potential to be developed further as a telehealth platform for elderly people in family, community health center (Puskesmas), and nursing-home settings.