Privacy-Preserving Machine Learning on Non-Co-Located Datasets Using Federated Learning
Jyoti L. Bangare, Nilesh P. Sable, Parikshit Narendra Mahalle, Gitanjali R. Shinde · 2024
Federated learning in machine learning (ML) allows several users to train a model without sharing raw data. This strategy may be very helpful for training models with several data sources, any of which may provide a security risk. This chapter discusses applying federated learning techniques to geographically distant (non-co-located) datasets. This overview will cover the key ideas and methodology, the pros and cons of this strategy, and the current treatments and best practices. Privacy protection is a critical concern in the Internet of Things (IoT) ecosystem because of the sensitive and private data that IoT devices collect. ML techniques are extensively used to overcome the privacy concerns in IoT devices. We analyze heterogeneous data, privacy, and security issues and propose preprocessing, encryption, and safe data aggregation to address them. Federated learning on non-co-located datasets can improve decision-making in healthcare, finance, and transportation and uncover dispersed data sources. Federated learning improves non-co-located datasets in this area. Federated learning on non-co-located datasets has several advantages. In the context of the IoT and Wireless Sensor Networks (WSNs), federated learning is important because it enables distributed learning and decision-making while protecting data privacy and minimizing connection costs. This chapter provides comprehensive and up-to-date research on federated learning on non-co-located datasets for ML and data analytics academics and practitioners. This chapter will also provide useful information, practical strategies, and recommendations.