Deep Learning-Based Skin Disease Detection Using EfficientNetB4: A Case Study on a Multiclass Image Dataset

One of the most prevalent health problems affecting people globally is skin disease, and early detection is essential to successful treatment and improved patient outcomes. Accurately diagnosing these conditions can be difficult, though, particularly since many skin diseases have similar visual characteristics and necessitate a high degree of clinical expertise. Deep learning has demonstrated promise in recent years in resolving these issues by automating the classification of medical images. In this work, we investigate the classification of six types of skin diseases: Benign, Malignant, Akne, Pigment, Ekzama, and Enfeksiyonel. By fine-tuning a model pre-trained on the ImageNet dataset, we achieve a classification accuracy of 93.2%, along with consistently high F1-scores and ROC AUC values between 0.99 and 1.00 across all categories. These results suggest that deep learning can play a valuable role in supporting dermatologists with fast and reliable diagnostic tools. This paper also examines the specific challenges involved in classifying certain diseases and discusses possible improvements for future work.