Abstract:
This study presents a deep learning-based approach for classifying Mpox disease from medical image data, aiming to enhance accuracy and speed in clinical diagnosis. A dataset of 3,574 images, categorized into Monkeypox, Chickenpox, Measles, and Normal classes, was compiled from public sources, primarily Kaggle. The data was split into training (80%), validation (10%), and testing (10%) sets. Preprocessing included contrast enhancement, resizing, normalization, noise reduction, and data augmentation through rotation, scaling, and flipping to improve image quality and model generalization. Four deep learning architectures—ResNet50, EfficientNetB3, CustomCNN, and Vision Transformer—were trained and tested using metrics like accuracy, precision, recall, and F1-score. ResNet50 had the highest accuracy (99.11%) and the lowest misclassification rate (0.89%). EfficientNetB3 came in second with an accuracy of 97.32%. CustomCNN and Vision Transformer didn't work as well, which could mean that there were problems with feature extraction or the data being used. The methodology made sure that preprocessing was done the same way every time, that data was spread out fairly, and that the model was thoroughly tested. The results show that pre-trained deep convolutional models are great for classifying Mpox because they give reliable and expandable diagnostic options. In places with few resources, this framework makes it possible to set up automated, real-time detection systems that could help doctors find diseases early, which would improve patient outcomes and cut down on delays in diagnosis.
