A convolutional neural network-long short-term memory deep learning classification framework for deepfake detection
Abstract
The credibility of digital media is being greatly undermined by the rise of deepfake (DF) technology. To help solve this issue, this paper proposes a hybrid spatiotemporal deep learning model that combines the ResNeXt50 convolutional backbone and the long short-term memory (LSTM) model to capture both frame-level spatial artifacts and temporal inconsistencies across video sequences. The novelty of this paper lies in the development of a lightweight convolutional neural network (CNN)-LSTM model that integrates selective fine-tuning of ResNeXt50 with mean-based temporal pooling, enabling improved generalization and computational efficiency for robust DF detection in real-world scenarios. The model was trained and evaluated on the WildDF dataset of 42,456 video sequences. Experimental results revealed that the proposed CNN-LSTM outperformed five baseline models, achieving 79.54% accuracy, 75.80% precision, 73.78% recall, and an F1-score of 74.77%. The proposed approach provides a practical framework for combining spatial and temporal information in DF detection while maintaining a lightweight architecture. This framework can serve as an intelligent classification component within an advanced intelligent classification system, supporting automated and reliable identification of manipulated visual content.
Keywords
Deep learning; Deepfake; Long short-term memory; ResNeXt50; Video sequences
Full Text:
PDFDOI: https://doi.org/10.11591/eei.v15i6.11675
Refbacks
- There are currently no refbacks.

This work is licensed under a Creative Commons Attribution-ShareAlike 4.0 International License.
Bulletin of Electrical Engineering and Informatics (BEEI)
ISSN: 2089-3191
,
e-ISSN: 2302-9285
This journal is published by the
Institute of Advanced Engineering and Science (IAES)
.