Performance Enhancement in Pre-Trained Deep Learning Models for Monkeypox Skin Lesions Identification Using Feature Selection Algorithms

Monkeypox is an uncommon viral infection leading to skin eruptions resembling smallpox. Recent monkeypox outbreaks demonstrate the persistent danger presented by this virus. Accurate and timely diagnosis of monkeypox is important for the effective treatment and prevention of outbreaks. In this study, we propose an integration of feature selection algorithms for the deep features approach for the classification of monkeypox skin lesions. The deep pre-trained models (ResNet50, GoogleNet, and InceptionNetV3) are fine-tuned at initial stage. After that, deep model-based features are extracted and filtered by the feature selection algorithms. Finally, the selected features are then classified using traditional classifiers. The obtained results show that the classification selected of deep features achieved high performance and outperformed the original version of the pre-trained model. The highest performance metrics belongs to the case of ResNet50-based features and Grey Wolf Optimization giving 96.8%, 95.3%, 98.0%, and 96.5% in terms of accuracy, precision, sensitivity, and F1-score, respectively.