Akademik Veri Yönetim Sistemi
4th International Conference on Artificial Intelligence and Applied Mathematics in Engineering, D. Jude Hemanth,Tuncay Yigit,Utku Kose,Ugur Guvenc, Editör, Springer Nature, Zürich, ss.1-11, 2023
The second most fatal disease in the world is cancer. Skin cancer is
one of the most common types of cancer and has been increasing rapidly in recent
years. The early diagnosis of this disease increases the chance of treatment dramatically. In this study, deep learning models are developed for skin cancer subtype detection including a standard convolutional neural network (CNN), VGG16,
Resnet50, MobileNet, and Xception. The parameters of the standard CNN model
are regularized using batch normalization, dropout, and L2-norm regularization.
The hyper-parameters of this model are optimized using grid search, in which
early stopping is used to optimize the number of epochs. For the rest of the models, transfer learning strategies are employed with and without fine-tuning as well
as re-training from scratch. Data augmentation is performed for increasing the
number of samples in the training set further. The performances of the models
are evaluated on a Kaggle dataset that is developed for binary classification of
skin images as malignant or benign. The best prediction accuracy of 87.88% is
achieved using ResNet50 as the convolutional neural network model, which is
re-trained from scratch and with data augmentation applied