Analisis Performa Pre-Trained Model Convolutional Neural Network Dalam Klasifikasi Kulit Wajah

Aras, Suhardi
Anam, Asyrofi
Jes, Billy
Sahar, Devid

Jurnal Teknologi Informasi dan Ilmu Komputer
Vol 13 No 3: Juni 2026

Klasifikasi jenis kulit wajah secara otomatis menjadi kebutuhan penting dalam mendukung pemilihan perawatan yang tepat dan objecktf. Pendekatan deep learning berbasis citra menawarkan solusi yang lebih akurat dibandingkan metode konvensional. Penelitian ini bertujuan untuk menganalisis dan membandingkan performa tiga arsitektur model pre-trained Convolutional Neural Network (CNN), yaitu EfficientNetB7, ResNet50, dan MobileNetV3 untuk klasifikasi lima kategori. Dataset terdiri dari 1950 gambar wajah yang telah diproses melalui tahapan preprocessing serta augmentation. Evaluasi dilakukan menggunakan metrik akurasi, precision, recall, F1-score, confusion matrix, serta waktu komputasi. Hasil pengujian menunjukkan bahwa EfficientNetB7 memberikan performa terbaik dengan mencapai akurasi tertinggi sebesar 93%, diikuti oleh ResNet50 sebesar 91%, dan MobileNetV3 sebesar 87%. Dengan demikian, arsitektur EfficientNetB7 menunjukkan kemampuan ekstraksi fitur visual yang lebih unggul dan berpotensi diterapkan pada sistem identifikasi kulit wajah berbasis sistem cerdas.   Abstract Automatic classification of facial skin types has become an important requirement in supporting the selection of appropriate and objective skincare treatments. A deep learning approach based on image analysis offers a more accurate solution compared to conventional methods. This study aims to analyze and compare the performance of three pre-trained Convolutional Neural Network (CNN) architectures, namely EfficientNetB7, ResNet50, and MobileNetV3, for the classification of five categories. The dataset consists of 1,950 facial images that have been processed through preprocessing and augmentation stages. The evaluation was conducted using accuracy, precision, recall, F1-score, confusion matrix, and computational time metrics. The experimental results show that EfficientNetB7 achieves the best performance with the highest accuracy of 93%, followed by ResNet50 at 91%, and MobileNetV3 at 87%. Therefore, the EfficientNetB7 architecture demonstrates superior visual feature extraction capability and has strong potential to be implemented in intelligent facial skin identification systems.