Skin Cancer Detection & Classification Using Machine Learning

Abstract

Our work topic is skin cancer detection and its type detect. For that we are using CNN (Conventional Neural Network) specially confusion matrix and auc-matrix. Which actually able to detect the accuracy. Achieving a high-level performance with accuracy we use HAM 10000 data for getting this. We also use some model for getting the perfect accuracy value. Our data set is vast but imbalanced for that we are using up-sampling method for balancing the data. Our data is imbalanced but able for performing. We are use VGG-19, Xception net model for getting the accuracy and 94.7% accuracy we get. We are using the provided data from International Skin Imaging Collaboration (ISIC) 2018. Skin is important for the neurological system. A growth or mass of abnormal tissue in the epidermal layer of the skin is called a skin cancer. And this can sometimes have catastrophic results. Early cancer and tumor detection allows for quicker treatment and increased survival rates. Digital imaging systems can detect tumors. There are times, though, when it just isn't enough. The capacity to segment pictures is essential in medicine. This might be used to identify cancerous skin cancers. However, there are a lot of barriers that prevent picture segmentation. The disappearing gradient is one of these problems. Which implies that deep convolutional neural network training may require more time and computational resources. We introduce a Deep Convolutional Neural Network (CNN) for completely automated skin tumor segmentation in digital imaging data to solve the vanishing gradient problem. The recommended procedure starts by classifying digital skin photos using Resnet-50 to evaluate whether or not a tumor is present. The levels of accuracy of the two CNN models were then compared. The VGG-19 structure and Resnet-50 encoder were then used instead. And so far, the outcomes have been very remarkable. This property permits the propagation of gradients to higher layers before they are attenuated to insignificant or zero levels. Only when digital imaging systems have undergone preprocessing and enhancement utilizing methods like rotation, zoom, horizontal and vertical shift, horizontal and vertical shear, and flipping are they employed in our model. We were able to identify and localize the tumor in the skin using our suggested model since it produced superior outcomes.