Application of Convolutional Neural Networks in Skin Disease Prediction: Accuracy and Efficiency in Dermatological Image Analysis

Abstract

The use of dermatological images and convolutional neural networks (CNNs) to predict skin diseases is one of the most promising applications of data science to improve the diagnosis and treatment of skin diseases. The aim of this work was to achieve maximum accuracy and efficiency in skin disease prediction using dermatological images and CNN models. Based on dermatological images, the ability of five CNN models to predict skin diseases was evaluated. The ResNet50, Inception V3, VGG-19, DenseNet201, and EfficientNet models were evaluated using the Kaggle HAM10000 (human against machine with 10000 training images) dataset. The metrics used were accuracy, recall, and F1 score. As a result, the study found that skin disease classification has variable performance. VGG-19 and DenseNet201 showed high values for accuracy, recall, and F1 score, with accuracy close to 98%. These models demonstrated an effective ability to identify and classify different types of skin diseases. In contrast, ResNet50 and Inception V3 obtained mixed results, while EfficientNet showed variable results in predicting skin diseases from dermatological images. Finally, the importance of choosing the right CNN model to predict skin diseases from dermatological images can be highlighted. VGG-19 and DenseNet201 performed well in classifying various skin diseases, which could be useful for developing dermatological diagnostic support systems.