SkinNet: An EnsembleNet Technique to Detect Skin Cancer Using Pre-Trained Models

Skin cancer is the most common cancer that causes death in both men and women. However, early detection can prevent the death rate. In this study, an ensemble-net technique to classify skin lesions as benign or malignant is proposed. Pre-trained convolutional neural networks (CNNs) architecture, InceptionV3, ResNet50, and DenseNet-121 are used as the base model after different experiments. All the experiments are conducted using the Nadam and Adam optimizers, with varying learning rates for each optimizer, and extensive pre-processing techniques. The dataset is collected from the ISIC 2018 dataset which contains images of two categories benign and malignant for training, validation, and testing sets. The proposed pipeline demonstrates high performance, with an accuracy of 98.59% on training, and 98% on testing compared to the other state-of-the-art methods for skin cancer disease prediction. The findings of this study are expected to be useful in cancer diagnosis research and potentially assist dermatologists in clinical decision-making.