Convolutional Neural Networks (CNNs) have shown promising performance in automating the feature selection process for classification of images, yet achieving optimal performance remains a complex task. This paper proposes an evolutionary approach to optimize the feature selection classification of skin cancer (benign versus malignant). By using the pre-trained VGG-16 model to extract image features and then employing Principal Component Analysis (PCA) + Genetic Algorithm (GA) for feature selection, followed by a classifier to monitor accuracy, this methodology aims to yield a greater classification accuracy on minimum number of features. Artificial neural network and random forest classifier were employed for classification. Accuracy of the classifiers was observed on both, i.e., features optimized with only PCA and features optimized with GA-driven PCA. Results of the artificial neural network classifier are compared with that of the ensemble classifier (random forest).