A Deep Learning and Explainable AI-Based Approach for the Classification of Discomycetes Species

Abstract

This study presents a novel approach for classifying Discomycetes species using deep learning and explainable artificial intelligence (XAI) techniques. The EfficientNet-B0 model achieved the highest performance, reaching 97% accuracy, a 97% F1-score, and a 99% AUC, making it the most effective model. MobileNetV3-L followed closely, with 96% accuracy, a 96% F1-score, and a 99% AUC, while ShuffleNet also showed strong results, reaching 95% accuracy and a 95% F1-score. In contrast, the EfficientNet-B4 model exhibited lower performance, achieving 89% accuracy, an 89% F1-score, and a 93% AUC. These results highlight the superior feature extraction and classification capabilities of EfficientNet-B0 and MobileNetV3-L for biological data. Explainable AI (XAI) techniques, including Grad-CAM and Score-CAM, enhanced the interpretability and transparency of model decisions. These methods offered insights into the internal decision-making processes of deep learning models, ensuring reliable classification results. This approach improves traditional taxonomy by advancing data processing and supporting accurate species differentiation. In the future, using larger datasets and more advanced AI models is recommended for biodiversity monitoring, ecosystem modeling, medical imaging, and bioinformatics. Beyond high classification performance, this study offers an ecologically meaningful approach by supporting biodiversity conservation and the accurate identification of fungal species. These findings contribute to developing more precise and reliable biological classification systems, setting new standards for AI-driven research in biological sciences.