Leveraging Transfer Learning and Ensemble Models for the Classification of Monkeypox and Other Skin Lesions

While the globe is still getting over the effects of the COVID-19 pandemic, the Monkeypox virus is posing a new threat. The Monkeypox virus is spreading, although not being as deadly or infectious as COVID-19. New cases are being reported every day in many different nations. To treat illnesses such as Monkeypox, early diagnosis and accurate identification of skin lesions are essential. In this study, we have evaluated deep learning models for categorizing Monkeypox lesions using two datasets: Monkeypox Skin Image Dataset (MSID) and Monkeypox Skin Lesion Dataset (MSLD), as well as a merged version of them. Alongside various deep learning models, including VGG16, ResNet50, EfficientNetB0, Xception, DenseNet121, and MobileNetV2, an ensemble of high-performing models was examined. The ensemble model consistently outperforms the individual models. It has achieved an accuracy of 91% on the MSID dataset, surpassing DenseNet121 and MobileNetV2, which both scored 88.31%. Similarly, on the MSLD dataset, the ensemble reached 81.90%, while MobileNetV2 and DenseNet121 achieved 77.62% and 73.57%, respectively. For the merged dataset, the ensemble performed best with an accuracy of 92%, followed by VGG16 at 84% and DenseNet121 at 83%. These findings indicate that the ensemble learning technique outperforms individual models across all datasets, recognizing Monkeypox lesions from skin images more accurately.