<![CDATA[This study aims to develop an automatic facial skin undertone classification system based on deep learning using facial images. The problem stems from the manual undertone identification process, which is subjective and potentially inconsistent when determining skin color categories. The method uses a Convolutional Neural Network (CNN) and compares two architectures, ResNet50 and MobileNetV2, for classifying warm, cool, and neutral skin undertones. The dataset combines secondary data from the Hugging Face platform and primary data obtained through direct facial image capture. The study follows the CRISP-DM process, including business understanding, data understanding, data preparation, modeling, evaluation, and deployment. ResNet50 achieved an accuracy of 89.3% and a weighted F1-score of 0.893, whereas MobileNetV2 achieved an accuracy of 68.3% and a weighted F1-score of 0.684. Thus, ResNet50 demonstrated better and more stable performance for facial skin undertone classification.]]>
Copyrights © 2026
