Leveraging Convolutional Neural Networks for the Classification of Five Imbalanced Skin Cancers

The worldwide prevalence of skin cancer problem is a full-size problem, occurring by way of environmental elements together with the consequences of world warming and climate trade. The importance of a well-timed prognosis for effective intervention is indisputable; but, cutting-edge diagnostic strategies are regularly complex, pricey, and inaccessible to many individuals. In this study, the researchers utilized Convolutional Neural Networks (CNN) to improve the early detection of skin conditions, addressing the challenges of dataset imbalance and manual identification by employing the MNIST: HAM10000 dataset, which contains 10,015 images across five distinct categories. By employing data augmentation and the synthetic minority oversampling technique (SMOTE), the researchers developed a CNN model to manage class imbalance. The CNN model carried out excellent overall performance, with training and testing accuracies of 99% and 95%. These consequences imply the model’s ability to as it should be, consistently classify numerous types of pores and skin cancer illnesses. We evaluated our model using other performance metrics like precision (0.847-0.990) and F1-scores (0.886-0.977), which confirmed the robustness of the model. By comparing other deep learning models, our model showed superiority in accuracy, loss, and consistency among the five classes. This confirmed that our proposed CNN model with SMOTE techniques shows effectiveness in the classification and identification of the type of skin cancer.