Hierarchical explainable deep ensemble approach for robust monkeypox classification
Biomedical Signal Processing and Control, cilt.127, 2026 (SCI-Expanded, Scopus)
- Yayın Türü: Makale / Tam Makale
- Cilt numarası: 127
- Basım Tarihi: 2026
- Doi Numarası: 10.1016/j.bspc.2026.111158
- Dergi Adı: Biomedical Signal Processing and Control
- Derginin Tarandığı İndeksler: Science Citation Index Expanded (SCI-EXPANDED), Scopus, Compendex, EMBASE
- Anahtar Kelimeler: Ensemble learning, Explainable artificial intelligence, Image classification, Monkeypox, Transfer learning
- Gazi Üniversitesi Adresli: Evet
Özet
In recent years, the use of image-based artificial intelligence methods for early and accurate diagnosis of monkeypox has been increasing. In this study, a more diverse, balanced, and complex dataset was created to overcome the limitations of the Monkeypox datasets widely used in the literature. Different open source datasets were combined, and class imbalance was minimized by supporting data preprocessing and data augmentation techniques. The dataset includes both remote and close-up views of the disease, thus presenting a more complex structure compared to other datasets. A two-stage ensemble learning method was developed in the modeling process. In the first stage, successful pretrained models were trained, and the best three models were selected. In the second stage, these models were combined with Multi-Layer Perceptron, XGBoost, and MALM supported stacking methods, respectively, to create stronger ensemble structures. The performance of the developed models, accuracy, precision, recall, F1 score, AUC, MCC, and BAS were evaluated with multi-dimensional metrics. In addition, the decision mechanisms of the model were made interpretable with SHAP analyses, thus increasing the reliability of the model outputs. In particular, the 0.9887 test accuracy, 0.9898 AUC, 0.9650 MCC, 0.9798 precision, 0.9700 recall, 0.9749 F1 score and 0.9887 BAS values achieved by the proposed model shows that the proposed methodology can be effectively used in clinical decision support systems. Specifically, we created a three-class (monkeypox, healthy, others) dataset, and our proposed approach demonstrated successful performance with robust performance metrics. These findings demonstrate both the effectiveness and robustness of our approach.