Answers are grounded in this paper and cite their sources.
Context As farming becomes increasingly data-driven, maintaining information privacy and security and protecting sensitive agricultural data is a shared responsibility among all stakeholders in the agricultural ecosystem. A distributed machine learning approach offers a way to address this issue, as it enables the management of sensitive data such as crop production, animal health, or economic yield, which may lead to higher productivity and better traceability across agricultural systems. Objective Although different studies have explored the application of distributed machine learning in the agricultural sector, the essential characteristics and overall findings of this approach remain scattered across the literature. To the best of our knowledge, no comprehensive attempt has been made to systematically synthesize its features, architectures, technical considerations, and associated challenges. This study conducts a systematic literature review to identify the distributed machine learning approaches implemented in agricultural systems as reported in the literature. Methods Using a 10-year publication range from 2015 to 2025, the review identified 1,864 articles, of which 41 were selected as primary studies. The selected studies were analyzed to identify the types of distributed machine learning based on communication approaches, as well as the adopted system architectures and governance approaches. Results and Conclusions Of the 41 articles, 24 were published in 2025, indicating that research on distributed learning in the agricultural sector has become increasingly important over the years. This study presents the types of distributed machine learning based on communication approaches, with Federated Learning identified as the most frequently studied in agriculture, appearing in 32 studies. The study also discusses the adopted system architecture and governance approaches, showing a strong connection between the two that affects the entire data flow. Since the client–server model was the most frequently mentioned architecture, evidence emerged that centralized authority was the dominant governance approach. However, we also find that client-server architecture applied consensus- and collaboration-based governance. It happened because the authors combined client-server with blockchain architecture, where to aggregate parameters from local nodes, the global node needs a consensus from all local nodes. In addition, other architectures such as P2P and hierarchical distributed learning were also identified, in which no single server holds the authority to control the system in a P2P topology. Thus, selecting an appropriate architecture is essential, as it influences the entire system, including its governance model. Significance Based on these findings, this study identifies substantial opportunities for further advancement in distributed machine learning that can be pursued by researchers and relevant stakeholders in the future. These insights contribute a foundational understanding that can guide future research and real-world implementation.
No takes yet. Share an insight, caveat, or question.
Krisnawijaya et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: