Title: Eye Gender Classification Using Transfer Learning (VGG16) on EyeDataset
Authors: Mohammed Sabri Al-madani, Mohammed Mahdi Jarour, Samy S. Abu-Naser
Volume: 10
Issue: 2
Pages: 35-41
Publication Date: 2026/02/28
Abstract:
Automated analysis of eye images can support a range of biomedical and computer vision applications. This paper presents a deep learning pipeline for binary classification of eye images into female eyes (0) and male eyes (1) using Eye Dataset. The proposed approach uses transfer learning with a pretrained VGG16 backbone, combined with standardized preprocessing, data augmentation, and supervised training using a fixed train/validation/test protocol. The best model is saved using checkpointing based on validation loss, then evaluated on a separated test set with class-wise and overall accuracy reporting. Experimental results demonstrate that transfer learning provides strong performance for eye-image gender classification under variations in image quality and illumination. The trained model achieved 65.67%: Test Accuracy overall test accuracy, with classwise performance detailed in the classification report.