An Explainable Machine Learning Framework for Crop Production Prediction and Recommendation System in Bangladesh

As an agricultural country growing through developing processes, it is necessary to have good crop yield predictions as well as good crop recommendations to ensure the long-term growth of agriculture in Bangladesh. In this study, we are presenting a machine learning-based system for crop production prediction and crop recommendation by implementing Decision Tree, Random Forest, and XGBoost models on the SPAS-DatasetBD. In addition to many valuable pieces of information for our analysis, the dataset included the area, production, season, minimum humidity, maximum humidity, minimum temperature, maximum temperature, district, and crop name. As a result of the comparison among the models implemented, the XGBoost model gave us the most suitable $\mathbf{R}^{2}$ value (0.9230) for crop production prediction, whereas the Random Forest and XGBoost models showed the most suitable performance for crop recommendation with precision and F1-score values of 0.94, respectively. To enhance the explainability of the results, we have incorporated the SHAP-based Explainable AI method into the developed framework. It is anticipated that the proposed framework will enable farmers and decision-makers to make informed decisions about crop choice, thus contributing to the sustainable development of agriculture in Bangladesh.