Enhancing Deepfake Video Detection: A Hybrid CNN-LSTM Approach
Deepanshu Singh, Prabhdeep Singh, Rahul Bhandari · 2024
Deepfake progress exhibits serious issue to the existing of digital data by generating hyper-realistic videos wich are fake. This work offers a combined Deep Learning (DL) model that utilises Convolutional Neural Networks (CNNs) along with Long Short-Term Memory Networks (LSTMs) to detect and analyse deepfakes videos. CNNs get spatial characteristics from different frames, meanwhile LSTMs showcases temporal connections among frames. This paper tackles the alarming problem of deepfake technique's danger to authenticity of digital media by suggesting a strong solution. This work presents combined DL model, which combines CNNs and LSTMs, captures spatial and temporal variations in video data accurately, allowing for exact deepfake identification. This model performs exceptionally well on actual and fake video datasets, with a precision of 0.67, F1 score of 0.80 and recall of 1.00. Using both spatial and temporal clues, this technique provides a strong defence against the spread of altered material in digital domains. Future endeavours will prioritise dataset extension and model refining to ensure greater applicability and accuracy for deepfake detection attempts.