Deepfake detection has become increasingly important in the digital age, where advanced technology can create highly realistic fake videos that can deceive viewers. This is particularly critical when it involves high-profile figures such as political leaders. This study focuses on the deepfake detection of videos featuring Donald Trump, a prominent figure whose public speeches and appearances provide data for analysis. Previous research in deepfake detection
has primarily concentrated on technical aspects, such as pixel-level inconsistencies and audio-visual synchronisation. However, there needs to be more exploration into individuals’ behavioural patterns to identify deepfakes. Behavioural analysis, which includes examining facial expressions, body language, and speech patterns, can offer a complementary approach to existing technical methods, providing a more robust framework for detection.
This thesis aims to investigate the effectiveness of behavioural analysis and corpus linguistics in detecting deepfakes of Donald Trump. By analysing specific behavioural traits and comparing them to known authentic footage, this study seeks to identify distinct markers that can indicate the presence of deepfakes.
The research also examines the potential for these markers to be generalised across different contexts and subjects. A comprehensive dataset of authentic and deepfake videos of Donald Trump was compiled. Parameters of the SCAnS method were employed to analyse behavioural features such as micro-expressions, gesture dynamics, and speech patterns.
The results demonstrated that specific behavioural indicators, such as inconsistencies in micro-expressions and speech anomalies, were significantly more prevalent in deepfake videos.
The detection approach achieved a high accuracy rate, indicating that behavioural analysis is a viable and practical approach to deepfake detection. Behavioural analysis can enhance the detection of deepfakes by providing additional layers of verification beyond technical discrepancies.
This study’s findings suggest incorporating behavioural analysis into deepfake detection frameworks can improve reliability and robustness, particularly for high-profile individuals like Donald Trump. Further research is recommended to explore the scalability of this approach to other subjects and more diverse datasets.
