Design and Analysis of an Improved Deep Ensemble Learning Model for Melanoma Skin Cancer Classification

Abstract

Due to global warming and the ultraviolet rays of the sun, skin diseases are spreading rapidly. If skin diseases are not treated during early stages, they can be dangerous to human life. Melanoma is a deadly type of skin cancer. Dermatologists find it difficult to diagnose melanoma skin cancer because of its complex structure. Significant human lives could be saved if melanoma cancer was diagnosed quickly and accurately. Expert dermatologists diagnose melanoma by examining the lesion's color images. It is very difficult to diagnose melanoma manually, and there is a risk of human error. Therefore, computer vision-based methods that diagnose melanoma disease correctly are offered, and there is little room for error as compared to manual diagnostic methods. In this research, we took images from multiple publicly available ISIC (International Skin Imaging Collaboration) data sets and developed a balance data set that has 10,500 images for training and testing. An ensemble of four convolution neural network (CNN) architectures (ResNet50, EfficientNet B6, InceptionV3, Xception) were utilized and trained on this dataset for classification of melanoma skin cancer. The experimental results of the proposed model show that it correctly classifies melanoma skin cancer. The proposed model gives satisfactory results as compared to other state-of-the-art methods.