Eye Movements as Indicators of Deception: A Machine Learning Approach
Loading...
Date
2025-05-25
Authors
Foucher, Valentin
de Leon Martinez, Santiago Jose
Moro, Robert
ORCID
Advisor
Referee
Mark
Journal Title
Journal ISSN
Volume Title
Publisher
ACM
Altmetrics
Abstract
Gaze may enhance the robustness of lie detectors, but remains under-studied. This study evaluated the efficacy of AI models (using fixations, saccades, blinks, and pupil size) for detecting deception in Concealed Information Tests across two datasets. The first, collected with Eyelink 1000, contains gaze data from a computerized experiment in which 87 participants revealed, concealed, or faked the value of a previously selected card. The second, collected with Pupil Neon, involved 37 participants performing a similar task but facing an experimenter. AI models (XGBoost) achieved accuracies of up to 74\% in a binary classification task (Revealing vs. Concealing) and 49\% in a more challenging three-classification task (Revealing vs. Concealing vs. Faking). Feature analysis identified saccade number, duration and amplitude along with maximum pupil size as the most important for deception prediction. These results demonstrate the feasibility of using gaze and AI to enhance lie detectors and encourage future research that may improve on this.
Gaze can enhance the robustness of lie detectors, but remains understudied. This study evaluated the effectiveness of AI models (using fixations, saccades, blinks, and pupil size) for detecting deception in hidden information tests in two datasets. The first, collected with the Eyelink 1000, contains gaze data from a computer experiment in which 87 participants revealed, concealed, or faked the value of a previously selected card. The second, collected with the Pupil Neon, included 37 participants performing a similar task but facing an experimenter. The AI models (XGBoost) achieved up to 74% accuracy in a binary classification task (Reveal vs. Hide) and 49% in a more challenging three-classification task (Reveal vs. Hide vs. Fake). Feature analysis identified saccade count, duration, and amplitude, along with maximum pupil size, as the most important for predicting deception. These results demonstrate the feasibility of using vision and artificial intelligence to improve lie detectors and will encourage future research that could improve this.
Gaze can enhance the robustness of lie detectors, but remains understudied. This study evaluated the effectiveness of AI models (using fixations, saccades, blinks, and pupil size) for detecting deception in hidden information tests in two datasets. The first, collected with the Eyelink 1000, contains gaze data from a computer experiment in which 87 participants revealed, concealed, or faked the value of a previously selected card. The second, collected with the Pupil Neon, included 37 participants performing a similar task but facing an experimenter. The AI models (XGBoost) achieved up to 74% accuracy in a binary classification task (Reveal vs. Hide) and 49% in a more challenging three-classification task (Reveal vs. Hide vs. Fake). Feature analysis identified saccade count, duration, and amplitude, along with maximum pupil size, as the most important for predicting deception. These results demonstrate the feasibility of using vision and artificial intelligence to improve lie detectors and will encourage future research that could improve this.
Description
Citation
ETRA '25: Proceedings of the 2025 Symposium on Eye Tracking Research and Applications. 2025, p. 1-7.
https://dl.acm.org/doi/full/10.1145/3715669.3723129
https://dl.acm.org/doi/full/10.1145/3715669.3723129
Document type
Peer-reviewed
Document version
Published version
Date of access to the full text
Language of document
en