Skin Disease Detection and Classification Using Machine Learning: A Comparative Study of CNN and SVM

Abstract

Millions of people are impacted by skin disorders every day around the world, and they are among the largest global health problems and place a great strain on healthcare systems. Traditional means of diagnosing skin diseases rely upon experienced dermatologists visually inspecting skin diseases, which can be subjective, very time consuming, and geographically limited to those who would be able to see a dermatologist in person. The introduction of machine learning and deep learning technologies has provided a way to develop automated systems for detecting skin diseases. The purpose of this paper is to compare two commonly used classification algorithms, Convolutional Neural Networks (CNN) and Support Vector Machines (SVM), for detecting and classifying four of the most common types of skin disorders: melanoma, eczema, psoriasis, and acne by using the ISIC 2020 dermoscopic images as the experimental dataset. The experimental results demonstrated that the CNN achieved an accuracy of 94.7% for classification where feature extraction was automatically done through deep convolutional layers, compared to a classification accuracy of 78.9% for the SVM where feature extraction was done using manually engineered features (Histogram of Oriented Gradients (HOG), Local Binary Patterns (LBP), and color histograms). In addition to comparing classifiers, the data preprocessing, augmentation techniques, and regularization used to train the classifiers will also be examined. While the SVM outperformed the CNN in classification accuracy, the CNN is a more computationally intensive classifier which makes it less desirable for deployment in resource-limited settings. Nevertheless, the SVM is a more interpretable classifier and is a good option to consider for deployment in resource-constrained settings.