Detecting Deepfakes With ResNext and LSTM: An Enhanced Feature Extraction and Classification Framework

With the growing availability of advanced editing tools and machine learning algorithms, it has become easier to create realistic and compelling deepfake videos, which are altered media content that is deliberately intended to deceive viewers. Deepfakes can have severe societal consequences, ranging from political propaganda to financial fraud, and they offer significant challenges for content authentication and fact-checking. This paper proposes an AI based technique which uses ResNext CNN and in combination with LSTM classifier to authenticate a video based on various visual features. The accuracy of this model is calculated at various sequence lengths for different datasets, and it is observed that the accuracy of this model rose with sequence length when testing it on various sequence lengths, and it almost begins to stabilize from the sequence length of 60.