Real-Time Deepfake Video Detection using Machine Learning: A CNN-based Authentication Framework

Deepfake technology presents significant threats to digital security, misinformation, and privacy. A Convolutional Neural network-based deepfake video detection in real time using optimization techniques to increase the accuracy and computational efficiency is proposed in this paper. The model is trained on the dataset of 15,000 images and uses feature extraction, adaptive loss function, and real-time processing optimization to achieve better performance. The experimental results give competitive results in terms of accuracy, precision, and recall compared to state-of-the-art models like EfficientNet and Xception. The framework is tolerant to adversarial attacks and compression effects. The work done in this direction will be further extended in the future to build transformer-based models for better generalization and adversarial resilience in deepfake detection.