A unified model of explainable regression and AES-based data security for soybean yield prediction

Accurate and interpretable crop yield forecasting is critical for enhancing agricultural productivity and ensuring food security. Soybean, a globally essential crop for protein, oil, and biofuel production, plays a pivotal role in food systems and agro-economies. Therefore, reliable yield prediction is crucial for strategic planning and resource allocation. This study proposes a comprehensive framework that integrates machine learning, explainable AI (XAI), and data encryption for robust and transparent soybean yield prediction. Twelve state-of-the-art regression models are systematically evaluated and optimized using Grid Search, Random Search, and Bayesian Optimization, with K-Nearest Neighbors (KNN) emerging as the top performer (R 2 = 0.8636). To enhance model transparency, eight XAI techniques including permutation importance, partial dependence plots (PDP), individual conditional expectation (ICE), SHAP, LIME, counterfactual explanations, residual analysis, and surrogate modeling are applied, highlighting ‘Cultivar’ and ‘NS’ as the most influential features. Furthermore, AES encryption in CBC mode is applied using AES-128, AES-192, and AES-256 to secure the dataset, with AES-256 demonstrating the highest confusion (50.02%) and diffusion (49.99%) while preserving decryption integrity. This unified approach delivers not only high prediction accuracy and interpretability but also ensures data security, establishing a strong foundation for secure, trustworthy, and scalable smart agriculture systems.