A Convolutional Neural Network Approach for Skin Lesion Classification Using Imbalanced Dataset with Image Augmentation

Abstract

A significant threat to people's health all over the world is skin cancer. The purpose of this research is to improve the detection of skin cancer by utilizing a CNN classification model that makes use of preprocessing and augmentation techniques. The HAM10000 dataset is used, and the imbalance it contains is addressed by resizing the images to 120×120 pixels and removing hair. Increasing the diversity of datasets through the use of data augmentation techniques is beneficial to the modeling and evaluation processes. In order to achieve the best possible classification of skin lesions, the proposed CNN architecture incorporates layers that have been carefully tuned. The data is divided into three different sets: Training, validation, and testing. The evaluation metrics, which include accuracy, precision, recall, and F1 score, all point to a highly successful performance of 0.932. This analysis demonstrates that the model is superior to other approaches to skin lesion classification, which signifies that it has the potential to be an effective instrument for the early detection of cancer.