Architecting Crop Recommendation Systems: A Conceptual Model of Hybrid and Machine Learning Approaches

Abstract

Machine learning in agriculture is an evolving field, with crop recommendation systems receiving significant research interest. A systematized synthesis of 51 journal and conference papers published between 2024 and 2026 identifies current trends, datasets, methodologies, and research gaps in the domain to support the development of a conceptual model. The analysis found that most existing systems use data from the same geographical location and are limited in crop diversity, which constrains their generalizability. Dataset diversity remains limited, with inconsistent reporting practices and varying transparency regarding data sources. Commonly used input features include environmental and soil variables such as temperature, rainfall, and macronutrient content, while micronutrients and physical soil properties are less frequently incorporated. Methodologically, recent trends indicate the adoption of explainable artificial intelligence (XAI), Internet of Things (IoT) integration, deep learning models, optimization techniques, and rule-based approaches. The approaches utilized are broadly classified into methodological hybrids, architectural integrations, and explainable integrations. Some studies incorporate complementary predictive tasks such as yield prediction and weather forecasting, while others develop integrated decision-support systems covering multiple stages of crop management. Based on these findings, the authors propose a flexible conceptual model. Overall, the findings highlight persistent challenges in dataset availability, model generalizability, and reproducibility. This study also identifies opportunities for developing more robust and transferable agricultural decision-support systems through dataset diversity, innovative hybrid and data-centric approaches, and consideration of usability for real-world adoption contexts.