Comparative Study of Crop Yield Prediction Using Explainable AI and Interpretable Machine Learning Techniques

Predicting crop yields is a vital aspect of agricultural planning as well as food security. For this, machine-learning models have become quite effective tools. Using a real-world dataset centered on agricultural production predictions, this research evaluates the interpretability and predictive outcomes of four renowned machine learning models: XGBoost, Decision Tree, Random Forest, and Linear Regression. Key indicators of performance involving Mean Absolute Error (MAE), R-squared (R2), and Root Mean Squared Error (RMSE) have been employed to gauge the models' ability to deliver precise predictions. XGBoost Regression consistently performed better than the other models, as shown by its lowest RMSE (1.46), MAE (0.68), and greatest R2 (0.95). The Decision Tree (98.33%), XGBoost (98.48%), Linear Regression (95.60%), and Random Forest (99.09%) models had the highest accuracy rates. To enhance the transparency and interpretability of the models, two explainable AI (XAI) techniques, SHAP (SHapley Additive exPlanations) and LIME (Local Interpretable Model-agnostic Explanations), were employed. These methods were utilized to provide deeper insights into the decision-making processes of the models. These techniques revealed the salient characteristics that shaped the models' forecasts. While LIME produced local explanations forecasts, SHAP values demonstrated the relative significance of factors in predicting crop output.