Klasifikasi Personal Color melalui Warna Kulit Wajah Menggunakan EfficientNetV2

Abstract

Determining clothing colors that align with individual charateristics, particulary facial skin tones, plays a significant role in enchancing self-confidence and aesthetic appearance. This study develops a web-based personal color classification system by facial skin color using the EfficientNetV2 model. Skin tones are classified into four primary categoriess: Light, Mid-Light, Mid-Dark, and Dark. The model was trained using a dataset with diverse skin color distributions to improve generalization and prediction accuracy. The test results show that the model achieved an accuracy of 94.8%, with average precision, recall, and F1-score values of 94.8%, 94.8%, and 94.8%, respectively. Despite the high accuracy, the system still experienced misclassification in classes with similar color characteristics, such as Mid-Light and Mid-Dark. These errors are mainly caused by variations in lighting, image quality, and uneven data distribution. The system also provides personalized color recommendations visually represented through images. Further development is necessary to enhance the model's performance, including expanding data diversity, employing more complex model architectures, and utilizing more complex model architectures. With these improvements, the system is expected to achieve higher accuracy and more consistent classification results.