Deteksi Masalah Kulit Wajah dan Rekomendasi Produk Berdasarkan Kandungan Perawatan menggunakan Vision Transformer

Abstract

Facial skin health plays a crucial role in supporting an individual’s self-confidence and quality of life. Various facial skin problems such as acne, blackheads, wrinkles, and hyperpigmentation are often difficult to detect independently without professional assistance, making it challenging to select the appropriate skincare products. With the advancement of artificial intelligence technology, deep learning-based image analysis offers a solution for building automated facial skin detection systems. This study develops a classification system for facial skin problems and a recommendation system for skincare products based on active ingredients using the Vision Transformer model. The facial image dataset undergoes preprocessing steps including resizing, normalization, and augmentation to improve input quality and increase data variability during training. The Vision Transformer model is fine-tuned on this dataset to accurately identify types of skin problems. Model performance evaluation using precision, recall, and F1-score metrics shows excellent results, each averaging 0.97 across five skin problem classes: acne scars, hyperpigmentation, acne, wrinkles, and blackheads. The model also achieves an overall accuracy of 97%, indicating a high capability in recognizing visual patterns in facial skin images. In addition to detection, the system applies a Content-Based Filtering (CBF) method to recommend active ingredients in skincare products that correspond to the classified skin problems. This system provides a practical and personalized solution to help users understand their skin condition and choose appropriate skincare products. Thus, this Vision Transformer -based detection and recommendation system serves as an effective innovative tool to enhance facial skincare literacy and assist users in making independent and accurate skincare decisions.