Deepfake Detection Using Convolutional Neural Networks: A Comparative Analysis

The rapid advancement of Artificial Intelligence (AI) has enabled the creation of hyper-realistic deepfake videos, raising serious concerns about misinformation, privacy breaches, and media manipulation. The research work introduces an innovative deepfake detection system utilizing a comparative analysis of nine state-of-the-art Convolutional Neural Network (CNN) models: MobileNet, NASNet, InceptionNet, InceptionResNet, XceptionNet, VGG16, AlexNet, LeNet, and EfficientNet. Through extensive experimentation on the DFDC dataset, VGG16 achieved the highest accuracy of 98.22%, with precision and recall scores of 98%, demonstrating its reliability in detecting even subtle manipulation artifacts. InceptionResNet and XceptionNet followed closely, with accuracies of 97.58% and 96.42%, respectively, highlighting the strength of hybrid and depthwise separable convolution architectures. The research also reveals that lighter models like MobileNet (86.87%) offer a tradeoff between efficiency and accuracy, making them suitable for mobile and resource-constrained environments. By integrating advanced data augmentation, frame-based analysis, and a systematic evalu- ation across multiple epochs, this work not only refines detection capabilities but also contributes to safeguarding digital content. The findings support global digital security efforts, aligning with the United Nations’ SDG 16 goal of promoting trustworthy information ecosystems.