Akhmetov, Adil2024-06-242024-06-242024-05Akhmetov, A. (2024). Federated Learning for Wearable Sensor Data. Nazarbayev University School of Engineering and Digital Scienceshttp://nur.nu.edu.kz/handle/123456789/7985Digitalization has revolutionized healthcare, with technologies like electronic records, apps, and wearables improving patient care. The combination of the Internet of Things (IoT) and Artificial Intelligence (AI) has raised healthcare standards, while the concept of Quantified Self (QS) encourages self-tracking and personal development. Yet, IoT data gathered for QS is not fully utilized due to analysis challenges, with only 7% of data properly used and interoperability issues complicating data exchange. Federated Learning (FL) provides a solution by enabling collective model training across different sources while protecting data privacy, in contrast to Centralized Learning (CL) which centralizes data. Overcoming interoperability is essential for effective data use, but complexities in digital health frameworks make this difficult. Thus, this work aims to investigate the use of FL, leading to the development of the web application that enables to conduct a predictive analysis of interoperable data from wearable devices and sensors in healthcare. The primary objectives of the work include: (1) Achieving data interoperability via FHIR protocol; (2) Evaluating the effectiveness of FL and CL in analyzing interoperable data; (3) Creating a user-friendly application using AutoML to train ML models on interoperable data in FL and CL approaches. In this work multiple datasets were analyzed by ML algorithms using CL and FL, and the results were evaluated using relevant metrics. Subsequently, a solution that leverages data interoperability is presented. The findings reveal that FL learning is more preferable than CL for preserving interoperable data confidentiality. The implementation and evaluation demonstrate the effectiveness of the web application in conducting the predictive analysis of centralized data.enAttribution-NonCommercial-ShareAlike 3.0 United StatesType of access: EmbargoMaster's thesis