A Study on Application of Model-Agnostic Explainable AI Techniques for Crop Recommendation

Sustainable agricultural practices play a major role in meeting global food security amid the rising impacts of climate change. Intelligent crop recommendation systems have become essential to assist farmers to make informed decisions in crop selection. Although Machine Learning (ML) models offer strong predictive capabilities, their black-box nature poses challenges in gaining trust and ensuring accountability. To address this gap, this study proposes a crop recommendation framework that combines tree-based ML classifiers with model-agnostic Explainable Artificial Intelligence (XAI) techniques to enhance transparency and interpretability. The system predicts suitable crops using key weather features such as temperature, rainfall, humidity, and soil nutrients (N, P, K, and pH). Among the models evaluated, Random Forest (RF) achieved the top accuracy of 99.55 percent and selected for interpretability and fairness analysis. Subsequently, the system adopts three model-agnostic XAI methods, SHapley Additive exPlanations (SHAP), Local Interpretable Model-Agnostic Explanations (LIME), and Partial Dependence Plots (PDP) to provide both local and global explanations of model behavior, focusing on the 'rice' crop. The result analysis demonstrated that rainfall, nitrogen, and humidity are the most influential features globally, with temperature consistently contributing in local instances. Fairness evaluation through SHAP summaries, LIME comparisons and counterfactual analysis demonstrated stable and justifiable predictions. Therefore, the proposed system shows strong potential for application in real-world agricultural decision support systems.