Machine learning-driven sensor fusion for precision agriculture: current applications and future perspectives

Purpose This study aims to provide a critical, comprehensive assessment of machine learning-driven sensor fusion as a core enabler of Agriculture 4.0. The review seeks to clarify how synergistic integration of diverse sensing modalities and advanced analytics can overcome the limitations of single-sensor systems, improve decision-making accuracy and enhance resilience under operational uncertainty. It also examines three linked barriers to adoption and offers recommendations to build trustworthy agricultural AI that farmers and advisors can understand and use fairly. Design/methodology/approach We conducted a structured literature search (Scopus, Web of Science Core Collection, IEEE Xplore and Google Scholar alerts) for 2005–June 2025 using Boolean strings combining “precision agriculture/agron*”, “sensor fusion/data fusion/multi-modal” and “machine learning/deep learning”. We applied inclusion criteria restricting to peer-reviewed English-language studies on agricultural applications of ML-enabled sensor/data fusion; reviews, editorials and non-agricultural domains were excluded. Screening proceeded at title/abstract and full-text levels by two reviewers with consensus resolution. Findings Evidence from multi-sensor case studies shows that fusion with machine learning consistently outperforms single-sensor baselines, delivering up to 30% gains in predictive accuracy and significantly improved fault tolerance. Feature-level fusion emerges as the most versatile approach, while decision-level fusion excels in robustness. The integration of emerging paradigms represents a viable path toward addressing current barriers to widespread adoption. Originality/value This review moves beyond descriptive surveys by providing a cross-cutting analysis that links methodological design choices with their agronomic, economic and governance implications. It uniquely frames sensor fusion adoption as a socio-technical challenge, integrating technical performance evaluation with policy and ethical considerations. The resulting synthesis offers both a rigorous evidence base and a strategic roadmap for researchers, developers and policymakers aiming to realize the full potential of intelligent, data-driven agricultural systems.