Table 1.

A Comparison of prior research on deepfake video recognition.

Research	DF Collections	CNN Attributes
A. Jaiswal [24].	FF++	Bidirectional recurrent neural networks (RNN) with DenseNet/ResNet50 are used to analyze the spatiotemporal properties of video streams.
P.Dongare [18]	Hollywood-2 Human Actions	It takes into consideration the deep-fake video's temporal irregularities. Inception-V3+LSTM
S. Lyu [25].	Closed Eyes in the Wild (CEW)	Used long-term recurrent convolution networks the frequency of eye blinking VGG16+LSTM + FC
A. Irtaza [14].	Fusion of datasets	Differences in facial structure, missing detail in the eyes and mouth, and a neutral network and logarithmic regression model
Nguyen [23].	Four major datasets	VGG-19 + Capsule Network
Hashmi [26].	DFDC whole dataset	CNN + LSTM Used facial landmarks and convolution features
Ganiyusufoglu [27].	FF++, Celeb-DF	Triplet architecture, Metric learning approach