DeepForgeryNet: Supplementary Results for a Hybrid CNN-LSTM Forgery Detection Model

A 13.5 KB supplementary document from 2026 detailing the performance results of the DeepForgeryNet detection framework. The file, authored by Aarti Sardhara and shared under a CC-BY-4.0 license, reports model evaluation metrics including accuracy, precision, recall, F1-score, and AUC from cross-dataset experiments.

Use Cases

• Benchmarking new forgery detection models against reported accuracy (95.1%), precision (94.6%), and recall (94.2%) metrics.
• Analyzing the effectiveness of artifact-aware preprocessing combined with spatial-contextual feature learning for detection reliability.
• Studying model generalization stability based on cross-dataset experiments with accuracy over 92%.
• Investigating detection challenges for very small manipulations or heavily compressed media as noted in the discussion.

Strengths

• Reports specific, high-performance metrics: 95.1% accuracy, 94.6% precision, 94.2% recall, 94.4% F1-score, and 0.98 AUC.
• Demonstrates generalization stability with cross-dataset accuracy over 92%.
• Released under a permissive CC-BY-4.0 license for reuse.

Limitations

• The dataset is a 13.5 KB DOCX file containing textual results, not the underlying training or evaluation data.
• Row count and column-level documentation for any potential tabular results are unknown.
• Description metadata is limited; actual data quality and structure require manual inspection after download.

Provenance

Source

figshare, author Aarti Sardhara.

Collection Method

Results from training an end-to-end model on publicly available benchmark datasets.

Time Range

2026

Freshness

Last updated 2026-05-08 05:48:13; freshness should be verified.

Geography

null