Classification of skin cancer using deep learning

The scarcity of medical resources is making skin cancer one of the most prevalent diseases on the globe. Prevention of skin cancer relies on accurate detection. Prompt detection of skin cancer is a challenge for dermatologists, and deep learning has found applications in both supervised and unsupervised learning. For object recognition and classification, Convolutional Neural Networks (CNN) are superior to all other types of networks. The MNIST: HAM10000 dataset contains seven types of skin lesions and one hundred and fifteen samples for the experiment. Preprocessing steps include autoencoder/decoder segmentation, dull razor sampling, and data cleaning. The model is trained using transfer learning methods like Resnet 50 and Dense Net 169 and obtained accuracy 88.4 and 64.9%, respectively. Whereas, enhance classification accuracy, with Xception approx. 96%. Using under sampling, DenseNet169 achieved respectable accuracy and f1-measure, in contrast to Resnet50's oversampling. The goal of expanding this study is to increase the accuracy of forecasts by adjusting certain parameters.