Truthnet: AI Powered Deepfake Detection Using a Hybrid LSTM–CNN Model

Abstract: Deepfake growth at an accelerating rate presents major threats to security, privacy, and digital media authenticity. Standard approaches to deepfake detection using convolutional neural networks (CNNs) are very good at detecting spatial artifacts but not at detecting temporal inconsistencies between video frames. To overcome this issue, we introduce a hybrid CNN-LSTM deepfake detection model that leverages the best of CNNs for spatial feature extraction with long short-term memory (LSTM) networks for learning temporal sequences. Our model was trained and tested on the celeb-df dataset, which is one of the hardest benchmarks for deepfake forensics. Experimental outcomes prove that the hybrid model outperforms single CNN and LSTM baselines in terms of better accuracy, precision, recall, and F1-score. Results prove the efficacy of combining spatial and temporal modelling for deepfake detection and emphasize the promise of the approach for multimedia forensics and security in real-world applications.

Keywords: Deepfake detection, Hybrid Convolutional Neural Network-Long Short-Term Memory (CNN-LSTM), Celeb-df, Multimedia forensics