Deepfake Detection: Demodulate Synthetic Videos Using Deep Learning Models

A deepfake detection system that uses machine learning (ML) and deep learning (DL) models to detect manipulated videos and images is presented in the study. Being aware of such synthetic content is crucial considering the emergence of deepfake technology, which might alter photos, videos, and audio for malevolent objectives including fraud, extortion, and disinformation. Deepfake technology has been applied to solve various real-time problems but is also exploited for unethical and illegal purposes. As a result, developing research and detection models is crucial to prevent its misuse. We proposed a CNN-LSTM hybrid model for analysis of cropped images to improve the performance of fake video detection. The suggested method focuses on identifying fake videos using the Celeb-DF dataset, which consists of 1203 videos (795 fake, 408 real). Moreover, the benefits and drawbacks of the various deepfake detection techniques are examined. The paper indicates potential improvements in model accuracy through more datasets and improved architectures, and it emphasizes the significance of sophisticated detection techniques to mitigate the negative consequences of deepfakes. With cropped video frames and deep learning techniques, the model's accuracy increased from 79.06% with the original dataset to <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{8 6. 8 2 \%}$</tex> with cropped videos.