<![CDATA[Deepfake is a form of facial manipulation in videos that utilizes artificial intelligence-based models to generate highly realistic visual content. The increasing spread of Deepfake has the potential to cause misinformation, manipulate public opinion, and enable the misuse of digital content, making reliable detection systems increasingly necessary. This study develops a face-based Deepfake video detection system by combining spatial and temporal analysis within a unified processing framework. ResNeXt is employed to extract visual facial characteristics from each frame, while Long Short-Term Memory (LSTM) is utilized to learn facial pattern changes across frames in a video sequence. The dataset used is sourced from FaceForensics++, consisting of 1000 original videos and 1000 Deepfake videos. All data are processed through frame extraction and face detection stages. Performance evaluation is conducted using accuracy, precision, recall, and F1-score metrics. The experimental results show that ResNeXt, as the baseline model, achieves 77.67% accuracy and 76.66% F1-score, while integrating LSTM improves system performance to 81.00% accuracy and 80.41% F1-score. These findings indicate that the utilization of temporal information contributes to improved stability and accuracy in face-based Deepfake video detection.]]>
Copyrights © 2026
