<![CDATA[Deepfakes are AI-generated media that enable highly realistic face replacement and other identity manipulations in video. Their rapid progress has intensified the challenge of distinguishing forged content from genuine recordings, necessitating effective and reliable detection techniques. This study develops a deepfake detection method based on Convolutional Neural Networks (CNNs) that learns discriminative visual patterns characteristic of manipulated videos, integrated with metadata extraction as a complementary signal. The proposed pipeline comprises dataset acquisition by scraping YouTube, frame extraction, data preprocessing, supervised labeling into genuine versus manipulated classes, and model training. The hybrid model was evaluated using a dataset of 287 genuine and 3 manipulated videos. Experimental results show that the integrated model achieved an accuracy of 99.65%, precision of 100%, recall of 66.67%, F1-score of 80.00%, and an Area Under Curve (AUC) of 1.0. These results demonstrate that combining metadata extraction with visual feature analysis is robust in minimizing false positives, making it highly relevant for enhancing the reliability of digital forensic investigations.]]>
Copyrights © 2026
