ADVANCED MACHINE LEARNING TECHNIQUES FOR ARRHYTHMIA CLASSIFICATION

Abstract

With the development of Internet-of-Things (IoT) applications, the concept of smart healthcare applications has gradually emerged to be the main factor in medicine. In fact, this raises the need to have a secure system that is efficient at the same time, due to the limited resources of IoT devices. Many different techniques have been developed and studied recently. For example, with centralized learning (CL), all data are collected and processed in one place. But many of these models are heavy and lead to an infringement of patient's privacy. Hence, a Federated Learning (FL) approach helps in developing global application without storing the data in centralized cloud. Therefore, in this thesis, the concept of CL and FL using a convolutional neural (CNN) network is performed to identify and classify arrhythmia, while taking into consideration the accuracy and simplicity in simulating a system model that would be used in medical devices. The MIT-BIH dataset was used in this work to test and validate the proposed approach and compare it to other methods in the literature.