Deep Learning and Explainable AI Approaches for Crop Disease Prediction: A Systematic Review and Future Directions

Abstract

Crop diseases are responsible for huge losses in agricultural yields and financial stability, which constitutes a very real threat to world food security. Reliable identification of illness and crop proper disease classification in the initial phase are thus of critical concern in the context of sustainable agriculture and increased yield control. Explainable Artificial Intelligence (XAI) and Deep Learning (DL) have been the breakthrough technologies of recent years in agriculture analytics, offering the capability of auto-detection of diseases from complex image data and increasing end-user interpretability. This systematic review accounts for 20 scientific publications from 2022 to 2025 that discuss DL and XAI-based models for multi-class and multi-label prediction of crop diseases from diverse image datasets. The review encompasses convolutional, recurrent, transformer-based, and hybrid neural structures employed in classifying plant diseases such as leaf blight, rust, mildew, and mosaic infections. The review further focuses on interpretability models that allow feature attribution interpretability, thereby making decision-making through AI practical for precision agriculture. Key challenges such as dataset imbalance, domain generalization, environmental variability, and model interpretability are covered comprehensively. The review consolidates previous advances, presenting insightful information on the ways in which DL and XAI, together, can enable intelligent, transparent, and sustainable disease management practices in today's agriculture.