Skripsi
KLASIFIKASI PENYAKIT KANKER KULIT BERDASARKAN CITRA DERMOSKOPIS MENGGUNAKAN VISION TRANSFORMER
Skin cancer is one of the most common types of cancer and requires early detection to increase the chances of successful treatment. This study aims to classify skin cancer diseases based on dermoscopic images using the Vision Transformer Base (ViT-Base) method. The research process begins with a data preparation stage, including data collection and splitting into training, validation, and testing sets. The preprocessing stage involves image resizing, handling class imbalance using weighted oversampling, data augmentation, and image normalization. Subsequently, the Vision Transformer Base model is trained using the preprocessed data and evaluated on the test set to assess its performance. The best model is obtained with training parameters consisting of 20 epochs, a batch size of 16, a dropout rate of 0.2, a weight decay of 0.001, and a learning rate of 5×10⁻⁵. The model achieves an accuracy of 90.12%, a precision of 82.83%, a recall of 77.97%, and an F1-score of 80.07%. These results indicate that the Vision Transformer Base is capable of delivering strong performance, and the application of appropriate data balancing techniques can improve skin cancer classification performance.