<![CDATA[Scabies is a contagious skin disease caused by Sarcoptes scabiei and remains a health issue in many countries, including Indonesia. Clinical diagnosis is often difficult due to symptom similarities with other skin conditions. This study evaluates the ConvNeXt architecture for scabies classification from skin images to support automated diagnostic systems. The dataset consisted of 273 images (86 scabies and 187 non-scabies), expanded through cropping to 819 images, divided into 488 scabies and 331 non-scabies images. ConvNeXt was trained with fine-tuning at stage 3, implemented using PyTorch on an A100 GPU with the AdamW optimizer. Evaluation used 70:15:15, 80:10:10, and 5-Fold Cross Validation schemes with accuracy, precision, recall, and F1-score metrics. The best configuration (learning rate 0.0001, batch size 32) achieved 95.97% testing accuracy. Overall, ConvNeXt reached 97.75% average accuracy and an F1-score near 0.98, demonstrating highly competitive performance that slightly exceeds established baselines like ResNet-50 (97.69% accuracy). This shows its potential as an accurate and practical solution for automated scabies diagnosis in resource limited healthcare settings.Key words: Automated Diagnosis, ConvNeXt, Deep Learning, Image Classification, Scabies.]]>
Copyrights © 2026
