Abstract
Skin disease is one of the most prevalent diseases around the world. To ease the complexity of conventional computational methods, deep learning techniques are adopted. By leveraging transfer learning and deep learning models, Convolution neural networks have been built, trained and evaluated. From the available data set of 10,082 using CycleGAN data augmentation a total of 39507 images are generated. One challenge faced during the implementation is mainly to improve the stability of CycleGAN during training phase. To analyze the performance of CycleGAN, three CNN models based on architecture of MobileNet, LeNet and modified LeNet are considered. A comparative analysis based on performance of three architectures is done and it is concluded that modified LeNet has outperformed with a higher accuracy of 95.03%.