Multi-Class Skin Disease Classification: A Study of Transfer Learning Strategies for Deep Learning Models

Skin diseases are a common health concern globally, requiring accurate diagnosis and treatment. Various techniques, including medication, topical treatments, phototherapy, and surgery, are employed for their cure. The detection of skin diseases has evolved significantly, with advancements in AI Technologies like image processing, machine learning, and deep learning. Image processing techniques extract meaningful features from images, while ML and DL algorithms classify diseases based on these features. In this work, we utilize a dataset of 23 classes, each containing 192 training images and 20 testing images. The dataset used here is the class-balanced dataset. We employ image processing techniques to enhance the dataset, followed by classification using DL models such as VGG19 and ResNet50. Our models achieve 83.69% and 82.82% accuracy, demonstrating their efficacy for classifying the skin disease images. This study contributes to the advancement of automation work of skin disease classification for diagnosis and treatment planning.