Melanoma Cancer Classification Using ResNet with Data Augmentation

Abstract

Melanoma skin cancer is cancer that difficult to detect. In this study, have been done melanoma cancer classification using Convolutional Neural Network (CNN). CNN is a class of Deep Neural Network (Deep Learning) and commonly used to analyzing images data. A lot of data used on CNN can greatly affect accuracy. In this study, the objective is to get best ResNet model for classifying melanoma cancer and normal skin images. The dataset that used is ISIC 2018. ResNet is used because the model winning the ILSVRC competition at 2015. ResNet architecture model that used are ResNet 50, 40, 25, 10 and 7 models. The architecture trained using data train that has been augmented and undersampling. The validation result on each model calculated using F1 Score. After validation and F1 Score result from the model obtained, the result compared each other to select the best model. The best architecture is ResNet 50 without augmentation that gives a validation accuracy of 0.83 and f1 score of 0.46.