<![CDATA[The rapid advancement of artificial intelligence, particularly in computer vision, has led to the proliferation of deepfake technology, which enables the creation of highly realistic synthetic facial images. This study proposes a deep learning-based approach for detecting real and fake faces using convolutional neural networks (CNN), specifically ResNet18, ResNet34, and ResNet50 architectures. The dataset used includes a public dataset from Kaggle (140K Real and Fake Faces) and a locally collected dataset to evaluate model generalization. Data preprocessing such as resizing, normalization, and augmentation were applied to improve robustness. Training employed transfer learning with fine-tuning over multiple epochs. Evaluation metrics included accuracy, precision, recall, F1-score, confusion matrix, and inference time. The results showed that ResNet50 achieved the highest validation accuracy of 94.1%, outperforming the other architectures. The integration of local datasets and data augmentation significantly improved classification performance. This model demonstrates strong potential for real-world deployment in digital security systems requiring deepfake detection.]]>
Copyrights © 2025
