DEEPFAKE DETECTION USING A HYBRID RESNEXT AND LSTM ARCHITECTURE

Abstract

Deepfakes pose a serious threat to media authenticity and public trust. This paper proposes a hybrid deep learning model combining ResNeXt and LSTM to detect deepfakes by capturing both spatial and temporal inconsistencies. ResNeXt extracts detailed frame-level features, while LSTM models temporal dependencies across video frames. Evaluated on benchmark datasets such as DFDC and Celeb-DF, the model achieves high accuracy and robust performance. The results confirm that integrating spatial and temporal features significantly improves deepfake detection, offering a reliable approach for video-based forensic analysis.