Enhancing Supply Chain Efficiency in Agribusiness through Machine Learning Algorithms

Abstract

Machine learning algorithms have emerged as a promising solution to inefficiency, climate-related risks, and escalating food demand across agribusiness supply chains. However, fragmented data, limited computing resources, and institutional inertia continue to impede large-scale adoption. Guided by the PRISMA protocol, a systematic review examined forty peer-reviewed studies published between 2020 and 2024 in major engineering and agricultural databases. The analysis reveals that decision trees and deep neural networks dominate yield and price forecasting tasks, consistently achieving accuracies above 90%. Deep reinforcement learning reduces logistics costs by up to onefifth through dynamic routing and inventory scheduling, while support vector machines enhance pest detection and soil diagnostics. Research activity is heavily concentrated in Asia, accounting for more than half of the identified literature, with particular emphasis on water-and fertiliser-efficient management practices in India and China. Persistent challenges include the opaque “black-box” nature of deep models, the substantial data and energy requirements of reinforcement learners, and the lack of collaboration between agronomists and data scientists. The review highlights the necessity for interoperable data standards, incentives that lower entry costs for smallholders, and hybrid, explainable models that balance accuracy with transparency. Complementary investments in IoT sensing infrastructure and workforce upskilling could further unlock value, potentially reducing post-harvest losses by nearly one-third and advancing global targets on food security and sustainable consumption. Overall, the evidence base provides a clear roadmap for building more equitable, efficient, and climate-resilient agrifood systems while underscoring the importance of addressing ethical and regional data gaps in future work.