AUTOMATED MULTI-CLASS SKIN DISEASE CLASSIFICATION VIA CONVOLUTIONAL NEURAL NETWORK IN MATLAB ENVIRONMENT

Abstract [English]

Skin diseases represent major global public health issues since they impact millions of people and need precise and prompt diagnosis to avoid serious health problems. Traditional diagnostic methods depend mostly on doctors' expertise and manual examinations, which results in subjective evaluations and inconsistent results with delayed treatments. Convolutional Neural Networks (CNNs) have become essential tools for medical image analysis in recent years because they can automatically extract hierarchical features and classify complex visual patterns. The paper details a complete methodology to classify skin diseases through training a deep CNN architecture with publicly available datasets of dermatological images. The model uses multiple convolutional and pooling layers before adding dense layers and a softmax classifier for the purpose of skin lesion categorization into different disease classes. We used data augmentation methods to improve generalization and handle class imbalance while employing cross-entropy loss with Adam optimizer during model training. The evaluation of the system included metrics like accuracy, precision, recall, and F1-score together with ROC-AUC. The experimental results proved that our CNN model reached superior classification accuracy compared to traditional machine learning methods, which depend on handcrafted features. The model demonstrated strong performance in differentiating between visually similar disease types according to the confusion matrix analysis. The study confirms how CNN effectively classifies dermatological images while showcasing its potential to serve as a dependable diagnostic tool in healthcare settings. The research identifies limitations such as dataset restrictions and the requirement for high-resolution images to achieve better accuracy. The next steps will concentrate on blending attention mechanisms with mobile deployment capabilities while broadening the range of skin conditions in order to improve both the scalability and clinical relevance of the model.