Comparative Analysis of Machine Learning and Deep Learning Models for Deepfake Detection: Insights from the Celeb-DF Dataset

We live in a digital age where reality is a maze of mirrors, where manipulation of data is at the tip of our fingers. The spread of misinformation and the unwanted spread of deepfakes have become prevalent. In times like these, there is a dire need to create a foundation of security that can be generated by accurate deepfake detection models. In order to conduct this research, we train 12 deepfake detection models using the Celeb-DF dataset, which consists of real and artificial celebrity videos. We then evaluate the results using five metrics: Accuracy, Precision, Recall, F1 Score, and ROC AUC Score. Among all the models the highest accuracy score of 94% was achieved by ResNet50. The AUC score of the models were impressive ranging from 89% -98%. The main objective of this paper is to use a thorough comparative study of a blend of deep learning and machine learning models in order to address privacy and confidentiality issues in the era of digital deception and build confidence in visual media.