A review of deepfake detection techniques and comparative
analysis with a ResNeXt + LSTM framework
Volume: 14 - Issue: 10 - Date: 01-10-2025
Approved ISSN: 2278-1412
Published Id: IJAECESTU459 | Page No.: 111-116
Author: Jaya Raj
Co- Author: Prof. Swati Khanve ,Prof. Nitya Khre
Abstract:-Deepfakes are AI-synthesized data particularly images and videos that look so
authentic, that they can easily deceive any common man. This study aims to build a hybrid
detection system with the help of ResNeXt-LSTM, where the ResNeXt-50 variant is used for
extracting spatial features which are fed in an LSTM network that finds temporal inconsistencies
across video frames. Conventional neural networks like CNN and LSTM’s alone have become
incapable for detecting deepfakes obtained from the latest generation techniques. The hybrid
approach combines the power of learning from both frame-level artifacts and sequential
anomalies. This model is trained on a dataset obtained from combining three different datasets
namely FaceForensics++, DFDC and Celeb-DF. Due to the combined strength of ResNeXt-50
and LSTM, and a versatile dataset, the model shows improved accuracy, efficiency and
generalization, which makes it suitable for deployment in real-time systems and potentially
adaptable to resource-constrained devices.
Key Words:-Deepfake Detection, ResNeXt, Long Short-Term Memory (LSTM), Convolutional Neural
Network (CNN), Recurrent Neural Network (RNN) Spatiotemporal Analysis, Frame-level feature extraction
Area:-Science & Technology
DOI Member: 165.40.460
DOI Member:
Preview This Article
