Authors:
K. S. S. Sreekrishna, S. Jasper Sofi, Balika J. Chelliah

Addresses:
Department of Computer Science and Engineering, Artificial Intelligence and Machine Learning, SRM Institute of Science and Technology, Ramapuram, Chennai, Tamil Nadu, India. Department of Computer Science and Engineering, SRM Institute of Science and Technology, Ramapuram, Chennai, Tamil Nadu, India.

Abstract:

Cassava (Manihot esculenta) is a critical food staple in most tropical areas, but its productivity is seriously under threat from numerous diseases. Traditional methods of diagnosing cassava diseases are cumbersome and require specialised expertise, underscoring the need for automated disease classification. This work presents an Explainable AI-based disease Prediction and Classification Framework that integrates metadata-augmented knowledge distillation to improve cassava disease detection. The proposed approach combines deep learning-guided image classification with metadata inputs (i.e., climate factors and soil characteristics) to motivate model accuracy. Additionally, Explainable AI (XAI) methods such as SHapley Additive Explanations (SHAP) and Gradient-weighted Class Activation Mapping (Grad-CAM) are employed to provide explainable insights into model predictions, promoting trust and transparency in decision-making. To overcome the limitations of using deep learning models in low-resource settings, knowledge distillation transfers knowledge from a high-performance deep neural network to a compact model. This leads to a computationally effective framework that can perform real-time inference on edge devices. The incorporation of metadata is rigorously tested to measure its effect on disease classification performance. An experimental evaluation using a cassava leaf image dataset and its corresponding metadata shows that the proposed framework significantly improves classification accuracy compared to traditional image-only models. 

Keywords: Explainable AI (XAI); Cassava Disease Classification; Knowledge Distillation; Machine Learning; Metadata Integration; Precision Agriculture; Cassava Green Mottle (CGM).

Received on: 13/12/2024, Revised on: 21/02/2025, Accepted on: 12/04/2025, Published on: 07/09/2025

DOI: 10.69888/FTSHSL.2025.000503

FMDB Transactions on Sustainable Health Science Letters, 2025 Vol. 3 No. 3, Pages: 179-191

• Views : 171
• Downloads : 12

Download PDF