Enhanced Age, Gender, Race Estimation Using Multi-task CNN

Abstract: This research on Enhanced Age, Gender, and Race Estimation Using multi-task CNN presents a comprehensive evaluation of a multi-task deep convolutional neural network (CNN) model designed to simultaneously estimate age, gender, and race from facial images. The testing utilizes real-world datasets, such as UTKFace, Adience, and MORPH II, along with a synthetic dataset that simulates ideal conditions (100% prediction accuracy) for baseline validation. The evaluation includes Mean Absolute Error (MAE) for age estimation, classification accuracy for gender and race, and one-off age accuracy to account for predictions in neighboring classes. Confusion matrices and distribution analysis provide deeper insights into the model's performance across different demographic groups. Although datasets such as UTKFace, Adience, and MORPH II present challenges due to variations in age, gender, and distributions, the proposed model demonstrates strong and high predictive accuracy. The results show that the proposed model surpasses state-of-the-art approaches, achieving an age estimation MAE of 2.95, gender classification accuracy of 98.3%, race classification accuracy of 93.1%, and one-off age accuracy of 90.7%. The addition of synthetic data proved beneficial in enhancing model robustness by mitigating demographic bias and improving prediction reliability. The findings of this study have practical implications for developing fair and reliable demographic estimation systems, with potential applications in security, human-computer interaction, and healthcare. Future work will focus on integrating attention mechanisms, fairness-aware learning, and domain adaptation techniques to enhance accuracy across diverse populations and uncontrolled environments.