A Systematic Literature Review of Distributed Machine Learning for Sustainable IoT Applications: Architectures, Privacy Preserving Techniques, and Green AI Metrics

Abstract

With the increasing deployment of IoT, the data is spread across the network, which is difficult to centralize because of privacy concerns, communication cost, and energy consumption. This paper is a systematic literature review of distributed machine learning (DML) for sustainable IoT, which includes architectures, privacy-preserving mechanisms, and Green AI metrics. We review 19 key works from 2017 to 2025 and report three patterns: (i) federated learning (FL) is the most prevalent privacy-preserving cross-device architecture, (ii) hybrid privacy defenses such as differential privacy and aggregation provide greater protection against inference and poisoning attacks, and (iii) energy-aware optimization and adaptive communication reduces the system footprint compared to cloud-centric baselines. The review suggests a three-dimensional taxonomy of architecture, privacy and sustainability, summarizes quantitative trade-offs and offers implementation guidelines for resource constrained IoT conditions.