Privacy preservation of EHR by data anonymization and federated learning for IoT based smart city application in...

This paper addresses privacy and security challenges in smart cities, particularly in IoT applications that process sensitive data in sectors like healthcare and agriculture. Traditional security solutions often fail to meet the energy and resource constraints of IoT devices, creating a significant gap in data protection. To address this, we propose a two-phase lightweight privacy-preserving approach. In the first phase, DIVersed and Anonymized Instances (DIVA) are used to anonymize and diversify sensitive data, ensuring privacy while enabling efficient processing. This approach mitigates the risks of identifying individuals through sensitive information. In the second phase, Federated Learning (FL) is employed, a decentralized machine learning technique that allows IoT devices to collaboratively train models without sharing raw data, thus preserving privacy. The proposed approach is evaluated based on information loss, execution time, and scalability, with experiments varying key parameters such as the number of Electronic Health Records (EHR) and the value of k. The results show that our method significantly reduces information loss, improves data utility, and maintains efficient execution, even as the number of devices and data points scale. This demonstrates the feasibility and effectiveness of our approach for privacy-preserving IoT applications in smart cities. Our solution ensures data privacy while enabling energy-efficient machine learning, making it suitable for IoT devices with limited resources. It provides a scalable, practical framework for applications in critical sectors like healthcare, where privacy and data utility are of utmost importance.