Table 6 Empirical analysis of different detection methods on various datasets
From: A novel approach for detecting deep fake videos using graph neural network
Refs. | Detection method | Dataset | Result |
|---|---|---|---|
Xia et al. [32] | MesoNet and a preprocessing module | FF++ | AUC = 0.974 |
Celeb-DF | AUC = 0.943 | ||
Agarwalet al. [7] | Temporal, behavioral biometric with CNN | FF++, DFDC-P and Celeb-DF | Less susceptible to counterattacks and generalizes effectively |
Hussain and Ibraheem [31] | CNNs in conjunction with the Jaya algorithm optimization | DFDC | Accuracy rates = 98.3% |
Celeb-DF | Accuracy rates = 97.6% | ||
Wodajo and Atnafu [16] | Convolutional Vision Transformer | DFDC | Accuracy = 91.5% AUC = 0.91 Loss value = 0.32 |
Kharbat et al. [5] | Region-Aware Temporal Filter (RATF) | FF++ and Celeb-DF | Outstanding performance |
Proposed Model | Multi Fusion between GNN and CNN | FF++  | Accuracy rate = 95.09% |
DFDC | Accuracy rate = 99.3% AUC = 0.96 | ||
Celeb-DF | Accuracy rate = 98.9% AUC = 0.98 |