AN XAI-ENABLED ENSEMBLE LEARNING APPROACH FOR SUGARCANE LEAF DISEASE CLASSIFICATION

Sreelakshmi KP Dr R. Parameswari AN XAI-ENABLED ENSEMBLE LEARNING APPROACH FOR SUGARCANE LEAF DISEASE CLASSIFICATION

Agriculture is highly vulnerable to crop diseases, which can severely reduce productivity if not detected at an early stage. In sugarcane cultivation, infections such as Bacterial Blight and Red Rot are major threats, often spreading rapidly and causing significant yield losses. Recent advances in deep learning have improved disease classification accuracy, but the lack of interpretability in most models limits their practical adoption by farmers and agronomists. To address this limitation, we present, an explainable ensemble framework for sugarcane leaf disease prediction. The backbone of the model combines a Tversky Indexive Projected Graph Neural Network (TIPGNN), which captures structural lesion patterns through graph representations, with a Gaussian Distributed Convolutional Neural Network (GDCNN), which extracts visual features such as texture and color. Predictions from both branches are fused in an ensemble layer optimized using Nesterov Accelerated Gradient to ensure robust and accurate classification. The distinctive feature of the proposed framework is the Explainability Layer, which integrates Grad-CAM to highlight affected regions, SHAP to compute feature contribution scores, and Graph Attention to identify influential nodes and edges within the graph. This dual focus on accuracy and interpretability makes the system both effective and trustworthy. Experiments on a sugarcane leaf dataset show that the model outperforms baseline CNN and GNN approaches in terms of accuracy, precision, recall, and F1-score. Furthermore, the generated explanations enhance transparency, allowing predictions to be validated by domain experts. The proposed framework thus bridges the gap between high performance and trust, offering a practical solution for smart agriculture applications.
06 01 2026 06 01 2026 74 85 10.58532/nbennurAIKEB1P1C10 https://iipseries.org/viewpaper.php?pid=8649&pt=an-xai-enabled-ensemble-learning-approach-for-sugarcane-leaf-disease-classification