MIoT-FL: A privacy-preserving federated learning architecture for securing ECG data in heart disease prediction

Medical Internet of Things (MIoT) devices are becoming more common in the healthcare field, and are being used by more and more people to talk to doctors directly on a regular basis. Machine learning algorithms that do not require the collection of patient data and its combination with other types of patient data that have already been collected enable MIoT devices to perform large-scale real-time patient monitoring and diagnostics, as well as offering personalised treatment opportunities. Federated learning (FL) is a different, more decentralised approach that still protects patients’ privacy and safety. This paper introduces a MIoT-based federated learning (MIoT-FL) system and its corresponding architecture, which is designed to protect electrocardiography data for cardiac disease prediction. The proposed architecture incorporates Secure Aggregation techniques and Differential Privacy protections, as well as Expedited Authentication, to ensure secure communication between MIOT devices. Experimental results demonstrate that the MIOT-FL system uses FL to improve predictive accuracy while reducing the risk of data breaches, inference attacks and model poisoning. Experimental evaluation of the MIT-BIH arrhythmia dataset demonstrates that the MIOT-FL system surpasses all prior FL systems in terms of privacy, classification accuracy and prevention of inference attacks. Classification accuracy scores were 84.92%, 89.32% and 98.21% when tested with ɛ ɛ = 1 , 2 and 4, respectively. This proposed MIOT-FL system provides a safe and private means of predicting heart disease. It also sets new standards for smart, privacy-focused healthcare solutions, making healthcare operations more efficient and data safer while ensuring compliance with legal standards