

<?xml version="1.0" encoding="utf-8"?>
<?xml-stylesheet type='text/xsl' href='/oai.xsl' ?>
<OAI-PMH xmlns="http://www.openarchives.org/OAI/2.0/"
	xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance"
	xsi:schemaLocation="http://www.openarchives.org/OAI/2.0/
	http://www.openarchives.org/OAI/2.0/OAI-PMH.xsd">
<responseDate>2026-07-07T11:19:27Z</responseDate>
<request verb="ListRecords" metadataPrefix="oai_dc" identifier="oai:publications.copernicus.org:acp30709">http:///OAfile/jere-1781962.</request>
	<ListRecords>
	<record>
	<header>
		<identifier>/OAfile/jere-1781962xsd</identifier>

		<datestamp>2026-01-01 </datestamp>
		<setSpec>acp</setSpec>
		</header>
		<metadata>
		<oai_dc:dc
			xmlns:oai_dc="http://www.openarchives.org/OAI/2.0/oai_dc/"
			xmlns:dc="http://purl.org/dc/elements/1.1/"
			xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance"
			xsi:schemaLocation="http://www.openarchives.org/OAI/2.0/oai_dc/
			http://www.openarchives.org/OAI/2.0/oai_dc.xsd">
			<dc:title><![CDATA[Deepfake Detection and Generalization: A Comparative Study of Deep Learning Architectures]]></dc:title>
			<dc:creator>*,Ali</dc:creator>
			<dc:description><![CDATA[Deepfake technology, which enables the creation of highly realistic synthetic face images using deep learning, poses significant threats to information integrity and digital trust. This study presents a comparative evaluation of deep learning architectures for image-based deepfake detection, with a particular focus on generalization across different data distributions. Five models were implemented and evaluated: MesoNet, ResNet-50, EfficientNet-B4, Xception, and an extended Xception model retrained with an additional DF40-based external dataset. Experiments were conducted on the 140k Real and Fake Faces dataset, which contains 100,000 training, 20,000 validation, and 20,000 test images, and were extended with 25,696 external training images. All models were trained using transfer learning where applicable, CrossEntropyLoss with label smoothing, AdamW optimization, cosine annealing, mixed precision, early stopping, and checkpoint selection based on validation loss. The original Xception model achieved 99.73% accuracy on the in-distribution test set, but its external test accuracy dropped to 52.74%, with a fake recall of only 6.04%. In contrast, the Xception+DF40 model reached 92.15% accuracy, 0.9828 AUC, and 84.50% fake recall on the external test set while maintaining 99.10% accuracy on the original test set. These results show that data diversity and out-of-distribution evaluation are critical for robust deepfake detection.]]></dc:description>
			<dc:date>2026-01-01</dc:date>
			<dc:type>Text</dc:type>
			<dc:format>application/pdf</dc:format>
			<dc:identifier>-</dc:identifier>
			<dc:identifier>-</dc:identifier>
			<dc:source>-</dc:source>
			<dc:language>eng</dc:language>
		</oai_dc:dc>
	</metadata>
	</record>
</ListRecords>
</OAI-PMH>

