Authors: Abhishek Dubey, Shivanshi Sahu, Shivansh Mishra, Tanvi Khetrapal, Sonali Patidar
Abstract: Environmental monitoring is essential for analyzing ecosystem behavior, forecasting natural hazards, and supporting sustainable resource use. With the rapid expansion of the Internet of Things (IoT) and distributed sensor networks, it has become possible to gather large-scale, real-time data. However, traditional centralized machine learning methods introduce significant chal-lenges related to data privacy, ownership, and regulatory requirements. Federated Learning (FL) has emerged as a decentralized approach that enables multiple nodes to collaboratively train mod-els without exchanging raw data, thereby protecting sensitive information. In this work, a Priva-cy-Preserving Federated Learning (PPFL) framework is introduced to provide secure, scalable, and efficient environmental data analysis across heterogeneous IoT systems. The proposed framework incorporates Differential Privacy (DP), Secure Aggregation, and Homomorphic En-cryption (HE) to ensure the protection of sensitive data during both communication and model updates. Experimental results show that the PPFL approach achieves a predictive accuracy of 97.6% with an RMSE of 0.203, surpassing recent FL-based techniques by up to 5.3%, while also maintaining strong privacy safeguards and reducing communication costs. This research presents a novel integration of differential privacy, homomorphic encryption, and edge computing within a unified federated learning framework for real-time environmental monitoring, effectively balanc-ing accuracy, privacy, and scalability.
DOI: http://doi.org/10.5281/zenodo.20179813
