Deep Learning Approaches for Age, Gender, and Ethnicity Prediction from Facial Images

Pimpale, Rohit Rajendra (2025) Deep Learning Approaches for Age, Gender, and Ethnicity Prediction from Facial Images. Masters thesis, Dublin, National College of Ireland.

Preview	PDF (Master of Science) Download (6MB) | Preview
Preview	PDF (Configuration Manual) Download (803kB) | Preview

Abstract

Predicting demographic traits like age, gender, and ethnicity directly from facial images underpins a wide range of tailored services, security mechanisms, and social data insights. Yet, the prediction task continues to be hindered by factors like lighting shifts, facial expression variations, ethnic class imbalance, and continuous age-related changes. To tackle these obstacles, the UTKFace dataset was leveraged, which includes around 24,000 facial photographs annotated across a broad spectrum of ages and ethnic backgrounds. Deep learning strategy was designed that integrates three modern convolutional architectures MobileNetV2, ResNet50, and EfficientNetB3 modified for simultaneous multi-output assignment. Each network learned to estimate continuous age and to classify gender and ethnicity at once. Enhanced by focused data augmentations and adjusted loss balancing, the lighter-weight MobileNetV2 recorded the best overall results, reaching a mean absolute error of 5.567 years for age, 91.81These findings affirm that compact architectures like MobileNetV2 can align high predictive accuracy with rapid inference (92.9 images/sec) and modest carbon footprint (0.091 kg CO2 emission). Grad-CAM was further employed to visualise attention maps, reinforcing the transparency of the models’ decisions across all outputs. Consequently, this study supplies a thorough benchmark of demographic predictors, underscoring the practical compromises of accuracy, efficiency, and explainability that must be navigated in scalable, real-world systems.

Item Type:	Thesis (Masters)
Supervisors:	Name Email Haque, Rejwanul UNSPECIFIED
Uncontrolled Keywords:	Age estimation; gender classification; ethnicity recognition; deep learning; convolutional neural networks (CNN); MobileNetV2; ResNet50; EfficientNetB3; UTKFace dataset; multi-output learning; facial analysis; Grad-CAM; model interpretability; real-time prediction; carbon footprint
Subjects:	Q Science > QA Mathematics > Electronic computers. Computer science T Technology > T Technology (General) > Information Technology > Electronic computers. Computer science Q Science > QA Mathematics > Electronic computers. Computer science > Computer Systems > Computers > Electronic data processing > Electronic data processing--Distributed processing > Edge computing T Technology > T Technology (General) > Information Technology > Electronic computers. Computer science > Computer Systems > Computers > Electronic data processing > Electronic data processing--Distributed processing > Edge computing H Social Sciences > HQ The family. Marriage. Woman > Gender
Divisions:	School of Computing > Master of Science in Data Analytics
Depositing User:	Ciara O'Brien
Date Deposited:	02 Jul 2026 14:55
Last Modified:	02 Jul 2026 14:55
URI:	https://norma.ncirl.ie/id/eprint/9448

Actions (login required)

View Item