EfficientResNetFusion: Hybrid Deep Learning Architecture with Multi-Method Explainability for Guava Fruit Disease Classification
Abstract
Guava diseases—Anthracnose and Fruit Fly—reduce marketable yield by 30–60%, yet automated fruit-specific classification remains underexplored and lacks principled class-imbalance handling and multi-method explainability. We propose EfficientResNetFusion, a dual-backbone hybrid CNN fusing EfficientNet-B0 (1,280-dim) and ResNet18 (512-dim) into a 1,792-dim fusion MLP head. LAB-space CLAHE preprocessing and pixel-space SMOTE balancing address contrast and class imbalance. The model is trained on the Kaggle Guava Disease Dataset (2,647 images; 3 classes) and externally validated on two independent datasets. Five XAI methods—Grad-CAM, SHAP, Saliency Maps, Integrated Gradients, and LIME—deliver multi-method interpretability. EfficientResNetFusion achieved 99.50% test accuracy, Macro F1 = 0.9942, MCC = 0.9924, Cohen’s Kappa = 0.9923, and AUC = 0.9999, surpassing all end-to-end baselines. SMOTE ablation confirms a +17.5 pp accuracy gain (82%→99.50%). Five-method XAI convergently confirms agronomically valid decision-making, establishing a reproducible, interpretable foundation for AI-driven guava disease management.
Keywords
References
- Ahmed, M., Ahmed, F., Naz, N. S., Mazhar, T., Khan, M. A., Khan, M. A., Ksibi, A., & Abbas, M. (2025). Automated guava disease detection using transfer learning with ResNet-101. Food Science & Nutrition, 13(12), e71348. [https://doi.org/10.1002/fsn3.71348]
- Ali, S., Abuhmed, T., El-Sappagh, S., Muhammad, K., Alonso-Moral, J. M., Confalonieri, R., Guidotti, R., Del Ser, J., Díaz-Rodríguez, N., & Herrera, F. (2023). Explainable artificial intelligence (XAI): What we know and what is left to attain trustworthy artificial intelligence. Information Fusion, 99, 101805. [https://doi.org/10.1016/j.inffus.2023.101805]
- Almutiry, O., Ayaz, M., Sadad, T., Lali, I. u. H., Mahmood, A., Hassan, N. U., & Dhahri, H. (2021). A novel framework for multi-classification of guava disease. Computers, Materials & Continua, 69(2), 1915–1930. [https://doi.org/10.32604/cmc.2021.017702]
- Atila, Ü., Uçar, M., Akyol, K., & Uçar, E. (2021). Plant leaf disease classification using EfficientNet deep learning model. Ecological Informatics, 61, 101182. [https://doi.org/10.1016/j.ecoinf.2020.101182]
- Azhar, N. A., Mohd Pozi, M. S., Mohamed Din, A., & Jatowt, A. (2023). An investigation of SMOTE based methods for imbalanced datasets with data complexity analysis. IEEE Transactions on Knowledge and Data Engineering, 35(7), 6651–6672. [https://doi.org/10.1109/TKDE.2022.3179381]
- Basak, S. K. (2024). Guava leaf disease dataset GLDD [external DS-2] [Retrieved June 30, 2025, from [https://www.kaggle.com/datasets/shuvokumarbasak4004/guava-leaf-disease-dataset-gldd]
- Farooqui, S., & Khan, Z. A. N. (2025). Deep learning driven disease diagnosis in guava leaves. International Journal of Engineering Research & Technology, 14(5). [https://doi.org/10.5281/zenodo.18106415]
- Galib, A. (2023). Guava disease dataset [primary, 3-class fruit] [Retrieved June 30, 2025, from [https://www.kaggle.com/datasets/asadullahgalib/guava-disease-dataset]
Details
Primary Language
English
Subjects
Artificial Intelligence (Other)
Journal Section
Research Article
Authors
Hamza Shahbaz
*
0009-0005-7552-1585
Pakistan
Publication Date
October 13, 2026
Submission Date
June 28, 2026
Acceptance Date
September 3, 2026
Published in Issue
Year 2026 Volume: 3 Number: 1