IMPLEMENTASI MODEL VISION TRANSFORMER PADA KLASIFIKASI JENIS KULIT WAJAH BERBASIS WEBSITE

Skin type misidentification often leads to inappropriate skincare product selection, which can negatively affect skin health. This study aims to develop a web-based automatic facial skin type classification system using the Vision Transformer (ViT) architecture. The model implemented is ViT Base Patch 16, pre-trained on the ImageNet dataset and fine-tuned using 10,000 facial images evenly distributed across four classes: normal, dry, oily, and combination. The dataset underwent augmentation and normalization during preprocessing. The training results showed an accuracy of 78% on the test data, with the best performance in the
combination skin class (F1-score of 0.86) and the lowest in the normal skin class (F1-score of 0.72). The model was integrated into a Flask-based system that enables users to classify their skin type by either uploading an image or capturing it via camera. System testing was conducted using functional testing and API testing via Postman. The results demonstrated that all key features of the system functioned
properly, and the API successfully returned classification responses in JSON format. This system can assist users in identifying their skin type and serve as a reference for selecting appropriate skincare ingredients.
Keywords: Classification, Self attention, Skin type, Transformer, Vision Transformer