A lightweight deep convolutional neural network model for skin cancer image classification

Deep learning models, particularly transformers and convolutional neural networks (CNNs), have been commonly used to achieve high classification accuracy for image data. Since introducing transformers, researchers have predominantly embraced these models to obtain impressive classification rates with novel approaches. In light of this scenario, we present a novel lightweight CNN called TurkerNet. Our primary objective is to attain a superior classification performance while minimizing the number of trainable parameters. TurkerNet comprises four essential components: the input block, residual bottleneck block, efficient block, and output block. To evaluate the performance of our proposed model, we conducted experiments using an open-access image dataset, specifically curated to include skin cancer images classified into two categories: benign and malignant. Our proposed (TurkerNet) model achieved a remarkable testing accuracy of 92.12% on this public dataset. Our model performed better than state-of-the-art techniques developed for automated skin cancer detection. Moreover, our proposed TurkerNet is a lightweight model. In this aspect, the presented TurkerNet is highly accurate with low trainable parameters.