Effect of Learning Rate on VGG19 Model Architecture for Human Skin Disease Classification

Abstrak

The skin is the largest external organ that serves to protect human internal organs and is very sensitive to various diseases, so early detection is very important to reduce the risk and increase the chance of recovery. This study aims to classify skin disease types using CNN algorithm with VGG19 architecture and learning rate adjustment to get a more optimal model, using a dataset from Kaggle consisting of 3,295 images with six classes, including several types of skin diseases and one healthy skin class. The preprocessing process includes dividing the data into training and testing sets, resizing the images to fit the VGG19 architecture, and normalization to scale the pixel values from 0-255 to a range of 0-1. The results show that using a learning rate of 0.00003 produces the best performance with 97.29% accuracy, 97.36% precision, 97.29% recall, and 97.30% F1-score. These findings confirm that the CNN algorithm with VGG19 architecture can classify skin disease types well.