Akademik Veri Yönetim Sistemi
5th INTERNATIONAL SELÇUK SCIENTIFIC RESEARCH CONGRESS, Konya, Türkiye, 14 - 15 Aralık 2024, ss.1700-1709, (Tam Metin Bildiri)
In the digital age, the proliferation of manipulated media content threatens public trust and information integrity. The increasing resemblance between fake content and real content is becoming increasingly apparent, as the methods of producing deceptive material diversify and the emergence of increasingly sophisticated and credible products continues to grow. In this context, the detection of deepfake videos has become a critical research area for digital security and integrity. The ResNet50 model used in this study aims to capture the fine details critical for deepfake detection by extracting important features from each frame of the videos. Celeb-Df was used as the dataset in the study. In the classification phase, two powerful machine learning algorithms, Random Forest and XGBoost, were chosen. This choice provides the opportunity to compare the performance of different approaches and determine the most effective method. Another noteworthy aspect of the study is the use of the AUC metric for performance evaluation. This choice allows for a more comprehensive evaluation of the performance of the classifier at different thresholds. AUC values of 93% were obtained with Random Forest and 95% with XGBoost. The high AUC values obtained show that the proposed method is quite successful in deepfake detection. The slightly higher performance of the XGBoost algorithm highlights its superiority in modeling complex data structures.