The New Advanced Society

14Skin Cancer Classification: Analysis of Different CNN Models via Classification Accuracy

Poonam Biswal, Monali Saha, Nishtha Jaiswal and Minakhi Rout*

School of Computer Engineering, Kalinga Institute of Industrial Technology Deemed to be University, Bhubaneswar, India

Abstract

Skin cancer is the most common form of cancer, which can be best treated if detected earlier. The main objective is to apply deep learning neural network in order to discover the relevant patterns which can help to guide the classification accurately. We have considered the well-known deep learning model that is convolutional neural network and compared the accuracy of the model by applying a vast dataset by varying the parameters such as number of layers, activation functions, etc. to find the best suitable parameters for CNN to design the classifier that could give the best accuracy while classifying the images of the seven types of skin cancer. We have achieved 84.31% accuracy in classification results. We have also mentioned models via Resnet application, where the CNN model gave an accuracy of 99.22%.

Keywords: Skin cancer, deep learning neural network, convolutional neural network, data loading, data processing, activation function, resnet, ROC, MSE curves

14.1 Introduction

Skin cancer is the most common of the cancer family to occur in humans. It can be classified into different types according to the region of the body where it occurs and physical characteristics of the type. It is a highly ...