Smart Farming using Machine Learning: A Review on Recent Trends and Approaches of Crop Yield Prediction

ABSTRACT Crop Yield Prediction (CYP) supports the finest decision‐making to support farmers for crop yields forecasting effectively which can help farmers make precise decisions about planting, harvesting, and development of their crops. However, CYP is a challenging endeavor due to the complexity of the underlying mechanisms and the impact of many variables, such as weather patterns, soil conditions, and crop management practices. Researchers have proposed different CYP models using Machine Learning (ML) techniques on different datasets and experimental settings. However, the reported performance of these models is often not directly comparable due to variations in data sources, spatial and temporal scales, crop types, and evaluation protocols. In this context, this review provides a conceptually integrated and analytical synthesis of CYP approaches. Rather than identifying a universally optimal model, this study highlights that model effectiveness is highly context‐dependent, varying with data modality, agro‐climatic conditions and application requirements. The article also discusses advantages and limitations of different ML models for CYP by a thorough analytical analysis. Besides, it highlights the key challenges for the implementation of ML models for CYP to produce improved crop productivity. This paper has the potential to be a helpful resource for both beginners and experts in the field of agriculture, particularly those who focus on ML‐based CYP and employing different varieties of crops. The outcomes of this review reveal that learning paradigms combined with multi‐source data integration deliver the most reliable yield predictions while also exposing critical gaps in crop diversity, climate coverage, and real‐world expert system deployment. This article is categorized under: Algorithmic Development > Ensemble Methods Technologies > Machine Learning Fundamental Concepts of Data and Knowledge > Explainable AI