Toward Explainable AI in Smart Permaculture: Design of the XCropSys Framework for Crop Recommendation

In the face of climate change, population growth, and resource scarcity, sustainable agriculture requires intelligent, adaptive, and transparent technologies. Permaculture offers a resilient model rooted in ecological principles, but its integration with digital systems remains limited. To support decision-making in sustainable agriculture, this paper proposes a Cyber-Physical System (CPS) architecture that integrates machine learning with local explainability for intelligent crop recommendation in permaculture contexts. The system processes environmental features such as soil nutrients, temperature, pH, and rainfall to predict suitable crops, while ensuring transparency through local explainability methods. We evaluated three classification models: Decision Tree, Random Forest, and Multi-Layer Perceptron, using a curated dataset of 2,200 labeled crop instances. The Random Forest model achieves the highest accuracy (99.55%) and is selected for further interpretability analysis. Local post-hoc explanations address explainability by highlighting feature-level contributions for each prediction, allowing human users to understand and validate AI-driven suggestions. The system architecture and experimental results do not include real-world CPS deployment; however, they illustrate how explainable AI can support decisions based on ecological principles in smart permaculture. This work contributes to the design of transparent, trustworthy, and domain-aware AI systems for sustainable agriculture.