ResNeXt-LSTM Hybrid Model for Deepfake Detection

Abstract

Deepfake technology has rapidly advanced, posing serious threats to digital media authenticity and security. To address this challenge, we propose a ResNeXt-LSTM Hybrid Model for efficient and accurate deepfake detection. The model integrates a ResNeXt-based Convolutional Neural Network (CNN) for spatial feature extraction and a Long Short-Term Memory (LSTM)-based Recurrent Neural Network (RNN) for temporal pattern analysis. The ResNeXt architecture enhances representational power through grouped convolutions, effectively capturing subtle facial inconsistencies across video frames. Meanwhile, the LSTM network models sequential dependencies to identify temporal anomalies introduced during video manipulation. Experimental results demonstrate that the proposed hybrid framework achieves superior performance compared to traditional CNN-only approaches, providing robust detection across diverse datasets and manipulation types. This architecture highlights the potential of combining deep spatial and temporal learning for next-generation multimedia forensics and deepfake mitigation systems.