An Explainable Hybrid Ensemble Framework for Region-Adaptive Crop Recommendation Using Multi-Source Agricultural Data

Abstract

Choosing an appropriate crop for a set of soil and environmental conditions exists one of most important decisions affecting agricultural productivity and long-term sustainability. Despite much of advances in machine learning-based agricultural analytics, larger existing crop recommendation models function as black-box systems and often face struggle to adapt across numerous geographic locations. Larger predictive performance may not always translate into practical considerations which farmers and agricultural planners can readily make use. In order to solve those challenges, the review gives an explainable hybrid ensemble system for region-adaptive crop recommendation which merges Random Forest, XGBoost, and Deep Neural Network models within a Bayesian optimization scheme. Proposed system integrates information from different agricultural sources, consisting soil nutrient properties, historical cultivation records, weather observations and satellite-derived vegetation indicators, permitting diverse environmental parameters to be simultaneously considered. The data quality was improved via missing-value imputation, dimensionality reduction, feature normalization and feature relevance analysis before the model training. The proposed framework was evaluated making utilizing agricultural datasets gathered from India and Africa under a 10-fold cross-validation strategy. The ensemble model performed an accuracy of 93.1%, an RMSE of 5.46, and an R² value of 0.95, outperforming the individual learning models. Evaluation on geographically different datasets further consistently demonstrated predictive capability, accuracy of 91.3% during cross-regional validation. Explainability analysis based on SHAP revealed that rainfall and vegetation-related indicators exerted the strongest influence on crop suitability predictions. The outcome demonstrates that integrating ensemble learning with explainable artificial intelligence can reliably provide transparent crop recommendations, offering support practically for resource-efficient farming, precision agriculture and climate-resilient agricultural planning.