Evaluating the Performance of Pre-Trained CNNs for Multi-Class Skin Disease Detection

This study evaluates the performance of pre-trained convolutional neural networks (CNNs) for multi-class skin disease detection. We investigate the efficacy of four models: Xception, InceptionResNetV2, ResNet50, and VGG16. The models are trained and tested on two publicly available datasets: DermNet (6 diseases) and a skin disease classification dataset. We compare the performance of each model based on metrics like accuracy, precision, recall, and F1-score. This research aims to identify the most suitable pre-trained CNN architecture for accurate skin disease classification using readily available datasets. The findings can contribute to the development of automated diagnostic tools to assist dermatologists in early and precise skin disease detection.