Deepfake Video Detection Using Hybrid CNN LSTM Architecture Integrating Spatial and Temporal Analysis
V N Manju, Aniket Ritam, S Supritha, H Jyothi · 2025
Deepfakes are generated by AI, which can alter the real content. The detection of deepfake images and videos is exploited using the Long Short-Term Memory (LSTM). LSTM belongs to the class of RNN, which efficiently captures the temporal dependencies that are present in sequential data, this makes it suitable for analyzing videos. The proposed system explores how LSTM architectures are effectively used in identifying deepfake videos, Moreover, it explains how the temporal patterns are used in altering the real content. The proposed model that helps in detecting deepfakes initially preprocesses the video data constructs training datasets that are of high quality and improves the generalization model using the augmentation techniques. The paper also discusses the different steps of training and optimization that are used exclusively by the LSTM networks for detection of the deepfake content and how they perform under various scenarios. Performance. Evaluation metrics such as accuracy, precision, recall, and F1 score are used to assess the model's ability to distinguish between authentic and manipulated content. The proposed methodology is compared with the existing methodology and shows a better performance with 96% accuracy.