Klasifikasi Tipe Dan Kondisi Kulit Wajah dengan Metode EfficientNet-B0 Menggunakan Grad-CAM sebagai Interpretasi Model

Abstract

Facial skin type and condition identification is an important step in determining appropriate skincare products. However, many people still face difficulties in recognizing their skin conditions independently, while direct consultation requires significant time and cost. This study aims to develop a classification system for facial skin types and conditions based on a Convolutional Neural Network using the EfficientNet-B0 architecture, as well as to apply Grad-CAM as a model interpretability method. The dataset used consists of 3,500 digital images categorized into seven classes, comprising skin types (acne-prone, oily, dry, and normal) and skin conditions (perioral dermatitis, aging, and vitiligo). The research stages include data collection, data preprocessing, data preparation, model training, model evaluation, and Grad-CAM implementation. The model was trained using a transfer learning approach with the Adam optimizer for 5 epochs. Initial testing with a learning rate of 0.0001 resulted in an accuracy of 89%. The system performance then improved significantly, achieving a final accuracy of 98% after adjusting the learning rate to 0.001. Grad-CAM visualization shows that the model focuses on facial regions relevant to each class. Therefore, the proposed method is capable of delivering high classification performance along with transparent visual interpretation.