DGAF-Net Research Portal Deepfake Detection in MIS

Conference-style summary

Deepfake Detection in MIS: DGAF-Net at a glance

A quick visual overview of the IEEE COMPAS 2025 study on DenseNet and multi-scale fusion for digital forensics in management information systems.

IEEE COMPAS 2025 FaceForensics++ DenseNet121 + Gabor Attention 88.33% accuracy
0% DGAF-Net accuracy
0 Reported AUC
0 Evaluated sequences
0 Training epochs

Problem framing

Deepfakes threaten trust in digital evidence, business communication, and operational decision-making in MIS. The study targets realistic media conditions with compression and manipulation noise.

Workflow

Workflow overview

DGAF-Net concept

DGAF-Net architecture

Dataset and training

  • 400-sequence balanced FaceForensics++ subset.
  • 1 fps frame extraction with ImageNet normalization.
  • AdamW, BCE loss, cosine annealing, 20 epochs.

Benchmark comparison

Model Accuracy AUC
ViT-B16 83.69% 0.84
Swin + CNN 85.00% 0.86
DGAF-Net 88.33% 0.88

Accuracy trend

Accuracy comparison

Analytics preview

Performance heatmap

Publication links

Full report Video IEEE Xplore

For full figures, the six derived plot assets, and reproducibility notes, continue to the report page or the repository README.