CNN-Based Model for Identity Recognition on Social Networks

Abstract

Identity recognition on social networks has become an essential technology for enhancing user authentication, preventing impersonation, and enabling personalised services, but it also raises concerns around privacy, bias, and security. As social platforms host billions of images and videos, the ability to accurately and fairly identify individuals in this vast multimedia space requires advanced machine learning techniques coupled with strong data protection measures. In this research, a secure, accurate, and fairness-aware facial identity recognition system is designed, implemented, and evaluated using Convolutional Neural Networks (CNN) as the core deep learning model. The system integrates robust security measures such as SHA-256 hashing and Fernet symmetric encryption to ensure end-to-end privacy protection in compliance with modern data regulations. Fairness in predictions was addressed using the Random Over Sampler technique to balance the dataset, resulting in equitable performance across simulated demographic groups, each achieving 0.97 in both accuracy and F1-score. The custom CNN architecture featuring multiple convolutional layers, batch normalization, dropout regularization, and dense layers was trained over 100 epochs on the widely used Labeled Faces on the Wild (LFW) dataset with augmentation, achieving a testing accuracy of 98.7%. The model’s accuracy consistently improved while loss decreased, confirming robust learning without overfitting. These results confirm the system’s superiority over traditional methods, offering a scalable, secure, and ethically aligned solution suitable for privacy-sensitive domains such as healthcare, online authentication, and secure access control.