A Review: A Robust Data-Driven Supervised Ensemble Machine Learning Framework for Soil Fertility Classification in Precision Agriculture Using Ensemble Boosting Models

Soil fertility is the foundation of sustainable crop production, and its timely, accurate assessment is central to the goals of precision agriculture. Traditional laboratory-based soil testing, while reliable, is slow, costly, and difficult to scale across the spatially and temporally variable conditions found in real farmland. Supervised machine learning, and ensemble boosting methods in particular, have emerged as a practical alternative capable of learning complex, non-linear relationships between soil physicochemical attributes (nitrogen, phosphorus, potassium, pH, organic carbon, electrical conductivity, and micronutrients) and fertility class labels. This review paper surveys recent data-driven approaches to soil fertility classification, with a specific focus on ensemble boosting algorithms including Gradient Boosting, AdaBoost, Extreme Gradient Boosting (XGBoost), Light Gradient Boosting Machine (LightGBM), and Categorical Boosting (CatBoost), alongside bagging-based baselines such as Random Forest and Extra Trees. The paper synthesises reported accuracies, methodological choices (data preprocessing, class-imbalance handling, feature engineering, hyperparameter optimisation, and model interpretability), and evaluation practices across a range of recent studies. Based on this synthesis, the review identifies persistent gaps, including limited dataset diversity, inconsistent benchmarking protocols, weak generalisation across geographies, and limited use of explainable AI, and proposes a conceptual robust ensemble framework intended to address them. The review is intended to serve as a foundation for the design of a data-driven supervised ensemble boosting framework for soil fertility classification in precision agriculture.