"Robust Model for Melanoma Classification Using Deep Learning and Machine Learning Techniques"

Abstract

"A major problem facing society today is skin cancer, particularly melanoma. This disease causes millions of deaths annually. Therefore, the aim of this research was to classify melanoma, also known as malignant mole, to dis-tinguish it and eventually detect it in time. The methodology employed con-tained six (06) phases: dataset acquisition, preprocessing (normalization, resizing), feature vector extraction (SIFT, HOG), Machine Learning models (RF, K-NN, SVM), Deep Learning models (EffientNetB7, NASNet, Vision Transformer (ViT), VGG-19), and evaluation (Precision, Accuracy, Recall and F1-Score). The results indicated that the ViT model performed best with an accuracy of 85.34%, followed by VGG-19 (83.78%) and NASNet (82.19%). In addition, the ViT model demonstrated high sensitivity with an AUC of 0.95. In conclusion, the implementation of artificial intelligence techniques can significantly improve the classification of melanoma, providing an objective and accurate tool compared to traditional visual inspection methods. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2025."