Explainable Artificial intelligence Model for Predictive Maintenance in Smart Agricultural Facilities

Artificial Intelligence (AI) applications in Smart Agricultural Facilities (SAF) often face limitations in terms of explainability, which can hinder effective adoption by farmers. To address this issue, the present study introduces a novel framework that integrates Predictive Maintenance (PdM) with eXplainable Artificial Intelligence (XAI). This model delivers predictive capabilities alongside interpretability across four critical dimensions: data, model, outcome, and end-user perspective. This integrated approach represents a paradigm shift in the way AI is applied and understood in the agricultural sector. The model demonstrates superior performance compared to existing approaches. Notably, the Long Short-Term Memory (LSTM) classifier achieves a 5.81% improvement in accuracy. Meanwhile, the eXtreme Gradient Boosting (XGBoost) classifier records a 7.09% increase in F1 score, a 10.66% boost in accuracy, and a 4.29% enhancement in the Receiver Operating Characteristic–Area Under the Curve (ROC-AUC). These advancements indicate the model’s strong potential for delivering accurate and reliable maintenance predictions in practical agricultural scenarios. Beyond predictive performance, the framework also provides in-depth insights into data integrity, global and local explainability, and counterfactual analysis relevant to PdM in SAF. By shifting the focus beyond conventional accuracy metrics, this study highlights the value of interpretability in AI-driven agriculture. The findings affirm the effectiveness of the proposed approach, making a significant contribution to the field. Additionally, the research encourages further exploration into the use of multi-modal data and Human-in-the-Loop (HITL) systems to enhance AI utility while addressing ethical dimensions such as Fairness, Accountability, and Transparency (FAT) in smart agricultural systems.