Skin Cancer Classification Malignant and Benign Using Convolutional Neural Network

Abstract

Skin cancer is one of the most deadly cancers. This cancer ranks third after cervical cancer and breast cancer in Indonesia. In detecting skin cancer, a dermatologist can carry out a biopsy. However, carrying out a biopsy requires a long time and preparation. Innovations to classify and detect skin cancer using artificial neural networks are overgrowing in helping doctors so that prompt and appropriate treatment can be carried out. The purpose of this project was to develop a system to classifying skin cancer using Convolutional Neural Networks (CNNs) and the ResNet50 architecture. This research examined the extent of system performance results using accuracy, recall, precision, and f1-score by doing several trials by changing the hyperparameters. The dataset used in this study was obtained online through Kaggle, with two classes, malignant and benign, divided into 80% training data and 20% test data. Based on the testing result, the best hyperparameter system was obtained using AdaMax optimizer, the learning rate was 0.0001, the batch size was 64, and the epoch was 50. In this research, The performance results values were 99% for precission, recall and f1-score. Simulation results show that this method with highly optimized hyperparameters can accurately classify malignant and benign skin cancer.