Data-Driven Crop Prediction: Combining Ensemble Methods with Interpretable AI

Choosing the right crop is a crucial decision that directly affects farm revenue, resource allocation, and risk management. Traditional practices rely on intuition and experience, which are often difficult to scale and adapt to evolving conditions. This paper introduces a streamlined pipeline that integrates a robust supervised baseline with systematic model analysis and interpretable inference for crop recommendation. A Random Forest classifier was trained on soil nutrients and local weather data to identify the most suitable crop. The model achieved 99.32% accuracy on the test set and consistent results across k-fold cross-validation (99.41% ± 0.37%). To ensure transparency and agronomic reliability, the pipeline incorporates global feature importance and SHAP-based explanations, enabling predictions to be both accurate and justifiable. The inference phase mirrors training, ensuring standardized inputs and facilitating integration into advisory tools. By combining ensemble methods with explainable AI, the framework provides precise and verifiable crop recommendations. Future work will focus on integrating additional data sources, incorporating risk-aware strategies, and supporting human-centered deployments to promote sustainable and reliable decision-making across diverse regions and seasons.