Classification of Skin Diseases to Detect Their Causes Using Convolutional Neural Networks

The cause of skin disease is a maj or factor in emerging skin diseases. Sometimes the cause of skin disease is still difficult to detect with only the sense of sight. To overcome this problem, deep learning methods such as Convolutional Neural Networks (CNN) are applied. In this work, several CNN architectures such as Inception V3, Xception, ResN et50, Seresnext101, GoogleNet, DenseNet121, EfficientNet are implemented to find out the best classification model. The causes of skin diseases used in this research are bacteria, fungi, infestation bite, and viruses, those skin diseases causes are obtained from the book “Skin Disease Diagnosis and Treatment”. The available image dataset requires several preprocesses techniques, including to remove the background, trim to maximum squared, dimension reduction, gray scaling, and normalization. The total 1004 images obtained from atlasdermatologico.com.br, dermatlas.net, dermnetnz.org, and hellenicdermatlas.com are used to train the model, and they are found unbalanced so that makes the model difficult to learn. Thus, a balancing process is required on the dataset. The evaluation result of the model is in the form of an accuracy metric. The evaluation results of each model will be compared, and then the best results will be selected which determines the best model. The top 3 model will be ensembled with an ensemble method, the result of this model will be compared to the best model.