2026-09-19T09:51:44Zhttps://dspace.mit.edu/server/oai/request

oai:dspace.mit.edu:1721.1/1629192025-10-07T04:13:20Zcom_1721.1_7582com_1721.1_7581col_1721.1_131023

Gupta, Amar Zen, Hilary Massachusetts Institute of Technology. Department of Electrical Engineering and Computer Science 2025-10-06T17:34:42Z 2025-10-06T17:34:42Z 2025-05 2025-06-23T14:04:42.733Z https://hdl.handle.net/1721.1/162919 Generation methods for deepfake images have advanced rapidly, and deepfake face images pose a critical security for biometric verification systems. Applications that rely on face recognition to grant access to sensitive data need to maintain high accuracy across a wide variety of deepfake generation methods, including novel and developing types that the application has not previously trained on. Current deepfake detection models achieve nearperfect accuracy on benchmark datasets, but do not perform as well on unseen types of deepfakes that were not part of their training dataset. We propose building an ensemble model with multiple base detectors, each trained on different generation model families to maintain high performance across many deepfake generation methods. Using four base models, including two models with the same architecture and training data, we exhaustively test all possible ensemble models. We find that combining similar base models trained on the same deepfake generation family does not improve performance compared to the individual base models. However, combining base models trained on different deepfake generation families leads to significant increases in accuracy and recall. Our ensemble framework provides a flexible and inexpensive solution in the ever-changing landscape of deepfake generation and security. M.Eng. Massachusetts Institute of Technology In Copyright - Educational Use Permitted Copyright retained by author(s) https://rightsstatements.org/page/InC-EDU/1.0/ Deepfake Face Detection: An Ensemble Framework for Generalized Classification in Biometric Verification Systems Thesis application/pdf Master Master of Engineering in Electrical Engineering and Computer Science Publication unknown unknown <Publication xmlns="https://www.openaire.eu/cerif-profile/1.1/" id="44711260-5761-4e87-9d6a-8552bc6ae10f"> <Type xmlns="https://www.openaire.eu/cerif-profile/vocab/COAR_Publication_Types">http://purl.org/coar/resource_type/c_1843</Type> <Title>Deepfake Face Detection: An Ensemble Framework for Generalized Classification in Biometric Verification Systems</Title> <PublishedIn> <Publication> </Publication> </PublishedIn> <PublicationDate>2025-05</PublicationDate> <Authors> <Author> <DisplayName>Zen, Hilary</DisplayName> <Affiliation> <OrgUnit> </OrgUnit> </Affiliation> </Author> </Authors> <Editors> </Editors> <Publishers> <Publisher> <DisplayName>Massachusetts Institute of Technology</DisplayName> <OrgUnit /> </Publisher> </Publishers> <License>https://rightsstatements.org/page/InC-EDU/1.0/</License> <Abstract>Generation methods for deepfake images have advanced rapidly, and deepfake face images pose a critical security for biometric verification systems. Applications that rely on face recognition to grant access to sensitive data need to maintain high accuracy across a wide variety of deepfake generation methods, including novel and developing types that the application has not previously trained on. Current deepfake detection models achieve nearperfect accuracy on benchmark datasets, but do not perform as well on unseen types of deepfakes that were not part of their training dataset. We propose building an ensemble model with multiple base detectors, each trained on different generation model families to maintain high performance across many deepfake generation methods. Using four base models, including two models with the same architecture and training data, we exhaustively test all possible ensemble models. We find that combining similar base models trained on the same deepfake generation family does not improve performance compared to the individual base models. However, combining base models trained on different deepfake generation families leads to significant increases in accuracy and recall. Our ensemble framework provides a flexible and inexpensive solution in the ever-changing landscape of deepfake generation and security.</Abstract> <Access xmlns="http://purl.org/coar/access_right" > </Access> </Publication>