Early Detection of Melanoma using EfficientNet B4 Transfer Learning Model

Melanoma is one of the most widespread types of skin cancer, and if not caught early, it can be dangerous. The number of avoidable fatalities can be decreased by using a reliable, automatic method that can detect the presence of melanoma from a dermatoscopic image of lesions and skin pigmentation. Deep learning algorithms have demonstrated promising results in recent studies for creating automated diagnosis for precise disease detection. This paper proposes an EfficientNetB4 network model architecture for organizing melanoma pictures into melanoma and not melanoma classes. The projected EfficientNet B4 model involves of 9 layers. In this proposed work, 17,805 images are used in the dataset. The model was trained 90 times, with learning rate 0.0001. Results shows that the proposed EfficientNet B4 model achieved a high accuracy of 96%. Overall, the study shows encouraging results and emphasizes EfficientNet B4's potential to be an essential tool for early detection of melanoma skin cancers.