Machine Learning Classifier Performance in Crop Recommendation Systems: A Systematic Review of Ensemble Versus Standalone Approaches

Abstract

Agriculture being the sector of prime importance which guarantees food security and contributes to the economy of a country, needs a robust and accurate crop growing model to achieve high yields. Advancement in the field of machine learning, has facilitated building recommendation systems that can recommend crops on various field and atmospheric attributes and contribute to real time prediction of crops. According to an analysis of research on machine learning algorithms, reported F1-scores, precision, and recall differ between ensemble and standalone classifiers. A stand-alone stochastic gradient descent classifier in one study obtained 100% F1-score, recall, and precision. Another achieved 99.2% precision, 99.4% recall, and 99.3% F1-score using an ensemble stacking technique (CaSR-Net). For standalone classifiers like random forest and severe gradient boosting, other study displayed precision values of roughly 98.5% to 99.6%; recall and F1 were not consistently reported. These results show that both standalone and ensemble methods work satisfactory when it comes to producing crop recommendations. The top-performing standalone and ensemble models displayed variations of approximately 0.5–1% across precision, recall, and F1-score when full metrics were available. In several studies, only partial metric data were reported, which limits the ability to draw broader quantitative conclusions.