Smart Agriculture: A Review of Machine And Deep Learning Techniques

Abstract

This systematic review evaluates the transformative role of machine learning (ML) and deep learning (DL) in advancing smart agriculture, focusing on precision farming applications. Using the PRISMA framework, we analyzed 127 peer-reviewed studies (92 journal articles, 30 conference papers, 5 book chapters) from Scopus, Web of Science, PubMed, and IEEE Xplore, published between 2015 and 2025. These studies highlight how ML and DL leverage agricultural big data to enhance crop management (yield prediction, disease detection, weed identification, crop quality assessment), livestock management (animal welfare, production optimization), and resource efficiency (water, soil, nutrient management). Notably, convolutional neural networks (CNNs) achieve over 98% accuracy in disease detection, while recurrent neural networks (RNNs) improve yield forecasting with R² values of 0.8–0.85. Challenges, including limited datasets, model generalization, computational complexity, and scalability, persist, particularly for smallholder farmers. Emerging solutions like transfer learning and federated learning address data scarcity and privacy concerns, enhancing accessibility. This review’s key contribution is a structured synthesis of ML and DL applications, identification of research gaps (e.g., imbalanced datasets, digital divide), and a proposed roadmap for scalable, sustainable farming innovations. It underscores ML and DL’s potential to drive data-driven, environmentally responsible agriculture, offering actionable insights for researchers, practitioners, and policymakers to advance precision farming.