A Hybrid Machine Learning Framework for Multi-Objective Optimization in Sustainable and Smart Agriculture

Machine learning (ML) is playing a key role in smart and sustainable agriculture by optimizing crop yields while maintaining long-term soil health. Though, most of the existing works address a few objectives in isolation, limiting their capacity to ensure better sustainability. To overcome the existing limitations, we propose a Hybrid ML Multi-Objective Optimization framework for Sustainable Crop Yield and Soil Health (HMLMOO-SCYSH). Our proposed framework integrates predictive modelling and evolutionary optimization to support data-driven decision-making in smart agriculture. Crop yield prediction is accomplished with help of XGBoost algorithm, while soil health is evaluated with help of a custom Soil Health Index (SHI) value predicted using Random Forest algorithm. Model interpretability and feature importance are guaranteed using SHapley Additive exPlanations (SHAP). The novelty of work lies in integrating SHAP-driven insights and Non-dominated Sorting Genetic Algorithm II (NSGA-II) for Pareto-optimal optimization of yield and soil health. Our framework was trained and validated on a real-time Indian agricultural dataset, achieving R² values of 0.75 for yield and 0.48 for soil health, indicating predictive performance. Investigational results disclose an optimal combination yielding 1030 kg/ha with an SHI of 16.17, demonstrating enhanced productivity with nominal soil degradation. In this article, we provides location-specific recommendations for farmers, assists policymakers in designing sustainable agro-input approaches, and contributes to the development of scalable Agri-AI systems for sustainable and smart agriculture.