Deep Learning Based Detection and Classification of Skin Diseases

Abstract

Skin diseases, ranging from acne to deadly melanoma, pose a major global health issue. Accurate, timely diagnosis is essential for effective treatment, but traditional methods are time-consuming and prone to human error. Advances in artificial intelligence, especially deep learning, now enable faster, more accurate dermatological diagnoses. This study explores deep learning techniques for classifying skin diseases using Convolutional Neural Networks (CNNs). Eight pre-trained models—including EfficientNet-B0, NASNet-Mobile, InceptionV3, ResNet variants, ShuffleNet, GoogleNet, and Inception_ResNet-v2—were evaluated using a Kaggle dataset. The focus was on classifying four key conditions (Acne, Rosacea, Eczema, Melanoma/Nevi) and identifying Urticaria. Models were assessed on accuracy, precision, recall, and F1 score. EfficientNet-B0 and ResNet-101 outperformed others, especially in detecting Melanoma/Nevi and Urticaria. While NASNet-Mobile showed lower accuracy, its lightweight design suits mobile use. Challenges remain, particularly in improving recall. The findings affirm deep learning’s value in dermatology and lay groundwork for AI-driven diagnostic systems. Future work will aim to boost generalization, mobile performance, and recall accuracy.