Enhancing Sustainable Agriculture with IoT-Based Soil Fertility and Crop Selection Using Hybrid Stacked Ensemble Models and Explainable AI

Sustainable agriculture depends on accurate soil fertility evaluation and data-driven crop choices. Still, conventional approaches have weaknesses in real-time analysis, scalability, and explainability. This study presents an IoT-ML-XAI architecture that combines IoT-based soil monitoring with a Hybrid Stacked Ensemble Model to facilitate precise and interpretable decision-making. The system utilises CNN, XGBoost, LSTM, and Random Forest for initial training, succeeded by a Neural Network meta-model for final predictions. XAI methodologies such as SHAP and LIME improve model transparency, enabling farmers to easily understand projections. The suggested approach surpasses conventional ML models, attaining 98.4% accuracy, 97.9% precision, and a decreased RMSE of 0.12, therefore substantially enhancing the dependability of soil evaluation. This framework decreases latency (1.2 seconds) and improves yield optimisation compared to traditional methods. The study enhances precision agriculture by integrating AI transparency with practical implementation. Future endeavours will concentrate on adaptive AI models and the incorporation of edge computing to improve practical applicability.