Gender Prediction Using Real-time Convolutional Neural Network

Balogun, Emmanuel O.; Ajagbe, Sunday Adeola; Adeniyi, Abidemi E.; Olayinka, Temitope O.; Adeogun, Elijah Adetoyese; Taiwo, Gbadegesin Adetayo; Esegbona-Isikeh, Ogheneruona Maria; Mudali, Pragasen

Abstract

This work presents real time gender prediction using Convolutional Neural Network. Automatic classification of gender has become an area that has garnered importance due to the emergence of breakthroughs in the world of computing particularly with the advent of machine learning and Artificial intelligence. The goal of gender prediction in computer vision involves accurate prediction of gender from visual data. Gender prediction is an indispensable biological metric as it plays a significant role in many human applications. The application varies from immigration, border access, law enforcement, defence and intelligence, citizen identification and banking. Image processing combined with machine learning algorithms have been employed to build solutions from image representation of biological attributes such as facial images, human skeletal radiographs (most notably skull and pelvic bones), gait, smiles and non-biological features such as social media activities, names and other means that could be employed to determine gender by extracting regions of interest, applying necessary filters and normalizing the matrix values obtained. For the purpose of this work, a total of 6760 hand-bone radiographs were acquired from the Radiological Society of North America Repository. The dataset was divided into 70% training datasets and 30% test dataset using Random Sampling Cross-Validation Method. The acquired data images were pre-processed using image cropping, histogram equalization and segmentation techniques. Performance evaluation was carried out on the developed system using the metrics: Accuracy, False Positive Rate (FPR), Recognition Time, Specificity and Sensitivity, these metrics were evaluated at threshold values of 0.25, 0.35, 0.5 and 0.75. Optimal values were gotten at optimum threshold value of 0.75, the values are; 4.79, 93.67, 95.21, 94.53 and 137.87 respectively for metrics (FPR, Sensitivity, Specificity, Accuracy and Time).

• Journal Article Type: Article
• Online Publication Date: May 10, 2025
• Publication Date: May 10, 2025
• Deposit Date: Jun 13, 2025
• Publicly Available Date: Jun 13, 2025
• Journal: Procedia Computer Science
• Print ISSN: 1877-0509
• Publisher: Elsevier
• Peer Reviewed: Peer Reviewed
• Volume: 258
• Pages: 497-506
• DOI: https://doi.org/10.1016/j.procs.2025.04.285