Gender Identification from Facial Images Using Custom Convolutional Neural Network Architecture

Abstract

Gender classification from facial images has become increasingly important in biometric applications. This study introduces a deep learning approach utilizing a custom convolutional neural network (CNN) model trained on 8,908 labeled facial images obtained from Kaggle, comprising 4,169 female and 4,739 male samples. Each image underwent preprocessing, including grayscale conversion, face alignment, cropping, resizing to 100×100 pixels, and pixel normalization. The CNN architecture consists of three convolutional layers with ReLU activation, max-pooling layers, a flatten layer, and two dense layers, ending with a sigmoid activation function for binary classification. The model was implemented using TensorFlow and trained for 70 epochs on Google Colab with GPU acceleration. Evaluation metrics include classification accuracy, confusion matrix, and area under the curve (AUC) from the ROC curve. The proposed system achieved 90.79% accuracy and 0.97 AUC, indicating robust performance. However, the confusion matrix revealed slightly higher precision for male predictions, suggesting the need for class balance refinement. The method demonstrates strong potential for integration into real-world facial analysis systems, such as identity verification, access control, and intelligent surveillance platforms.