Comparative Analysis of CNN and CNN-SVM Methods For Classification Types of Human Skin Disease

Abstract

Cancer is one of the leading causes of death worldwide, with skin cancer ranking fifth. The skin, as the outermost organ of the body, is susceptible to various diseases, and accurate diagnosis is crucial for effective treatment. However, limited access to dermatologists and expensive skin biopsies poses challenges in achieving efficient diagnosis. Therefore, it is important to develop a system that can assist in efficiently classifying skin diseases to overcome these limitations. In the field of skin disease classification, Machine Learning and Deep Learning methods, especially Convolutional Neural Network (CNN), have demonstrated high accuracy in medical image classification. CNN's advantage lies in its ability to automatically and deeply extract features from skin images. The combination of CNN and Support Vector Machine (SVM) offers an interesting approach, with CNN used for feature extraction and SVM as the classification algorithm. This research compares two classification methods: CNN with MobileNet architecture and CNN-SVM with various kernel types to classify human skin diseases. The dataset consists of seven classes of skin diseases with a total of 21.000 images. The results of the CNN classification show an accuracy of 93.47%, with high precision, recall, and F1-score, at 93.55%, 93.74%, and 93.62%, respectively. Meanwhile, the CNN-SVM model with "poly," "rbf," "linear," and "sigmoid" kernels exhibits varied performances. Overall, the CNN-SVM model performs lower than the CNN model. The findings offer insights for medical image analysis and skin disease classification research. Researchers can enhance CNN-SVM model performance with varied kernel types and techniques for complex feature representations.