<![CDATA[Acne vulgaris is a common dermatological condition that significantly impacts psychosocial well-being, particularly among adolescents and young adults. Accurate identification of acne lesion types is crucial for effective treatment planning, yet manual assessment by dermatologists is subjective and resource-intensive. This study proposes a Convolutional Neural Network (CNN)-based approach using EfficientNetV2-S with transfer learning and data augmentation to perform multi-class classification of five acne lesion types: blackheads, whiteheads, papules, pustules, and cysts. The model was trained and evaluated on 4,673 annotated facial images, achieving an accuracy of 96.66%, outperforming conventional lightweight CNNs and achieving comparable results to heavier ensemble architectures. Statistical validation using p-values and effect sizes confirms the model’s robustness. The scientific contribution of this research lies in the integration of EfficientNetV2-S with a customized classification head optimized for multi-class acne recognition—an area underexplored in dermatological AI research. Unlike previous works focusing on binary classification or ensemble models, our approach offers a lightweight, accurate, and scalable solution for real-world teledermatology, thus establishing a novel benchmark in multi-class acne classification.]]>
