Nowadays, computer vision plays an essential role in disease detection, computer-aided
diagnosis, and patient risk identification. This is especially true for skin cancer, which can be fatal if not
diagnosed in its early stages. For this purpose, several computer-aided diagnostic and detection systems
have been created in the past. They were limited in their performance because of the complicated visual
characteristics of skin lesion images, which included inhomogeneous features and hazy borders. In this paper,
we proposed two methods for detecting and classifying dermoscopic images into benign and malignant
tumors. The first method is using k-nearest neighbor (KNN) as classifier when pretrained deep neural
networks are used as feature extractors. The second one is AlexNet with grey wolf optimizer, that optimizes
AlexNet’s hyperparameters to get the best results. We also tested two approaches in classifying skin cancer
images, which are machine learning (ML) and deep learning (DL). The used methods in ML approach are
artificial neural network, KNN, support vector machine, Naïve Bayes, and decision tree. The DL approach
that we used contains convolutional neural network and pretrained DL networks: AlexNet, VGG-16, VGG19, EfficientNet-b0, ResNet-18, ResNet-50, ResNet-101, DenseNet-201, Inception-v3, and MobileNet-v2.
Our experiments are trained and tested on 4000 images from the ISIC archive dataset. The outcomes showed
that the proposed methods outperformed the other tested approaches. Accuracy of first proposed method
exceeded 99% in some models and second proposed method achieved 99%.