<![CDATA[Facial skin type determination is a critical aspect in the development of digital skincare recommendation systems and technology-based dermatological services. However, skin identification methods that rely on manual assessment still present several limitations, as they are subjective, time-consuming, and highly dependent on the observer’s expertise. To address these issues, this study proposes an automated approach for facial skin type classification using Convolutional Neural Networks (CNNs) with a transfer learning scheme. Three computationally efficient CNN architectures MobileNet, MobileNetV2, and MobileNetV3-Large were employed and evaluated comparatively. This study utilized a dataset consisting of 2,250 facial images categorized into five skin types, namely Combination, Dry, Normal, Oily, and Sensitive. The dataset was divided into 1,800 images for training, 225 images for validation, and 225 images for testing to ensure objective performance evaluation. All images were normalized and resized to 224 × 224 pixels prior to model processing. Model training was conducted using two epoch configurations, specifically 5, 10, 20, 30, 50 and 100 epochs, to examine the effect of training duration on classification performance. The experimental results indicate that increasing the number of training epochs has a positive impact on the accuracy of all evaluated models. Among the three architectures, MobileNetV3-Large achieved the best performance, attaining a test accuracy of 100% at 100 epochs and demonstrating superior generalization capability, particularly in distinguishing skin types with similar visual characteristics, such as Sensitive and Combination. These findings confirm that appropriate CNN architecture selection and training configuration play a crucial role in enhancing facial skin type classification performance and highlight the potential applicability of the proposed approach in mobile-based applications.]]>
