Feature Extraction for Facial Skin Type Classification Using Photo Images

Understanding skin type is the first step in selecting skincare or makeup products. This is done to choose appropriate skincare products, address specific skin issues, or minimize mistakes in selecting skincare treatments. Therefore, research needs to be conducted to develop a model aimed at classifying facial skin types using photographic images. In previous studies, image classification modeling using the CNN method has been shown to produce better results compared to other models. The modeling uses the Convolutional Neural Network (CNN) method by exploring its architectures, namely MobileNetV1, MobileNetV2, and MobileNetV3Large, to find the best architecture. A total of 588 images were successfully collected and used in the research, obtained directly and labeled by a doctor based on skin type. From this data, 80% was used for training and 20% for testing. The modeling process began with image data preprocessing, followed by training, testing, and model evaluation. The comparison of modeling results for each architecture showed that the MobileNetV3Large architecture was quite suitable for classifying facial skin types, with a validation accuracy of 45% and a training accuracy of 49%. After evaluation, it was concluded that the image data needed improvement. Based on a re-modeling with improved images using the MobileNetV3Large architecture, the results showed a validation accuracy of 48% and a training accuracy of 51%, indicating an improvement after the image data was refined.