Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Eye Movements as Indicators of Deception: A Machine Learning Approach

Valentin Foucher, Santiago de Leon-Martinez, Robert Moro

TL;DR

This study investigates whether eye movements can serve as indicators of deception within Concealed Information Tests (CIT). It evaluates gaze and pupil features from two datasets—Eyelink 1000 (computer-based) and Pupil Neon (wearable)—using an XGBoost classifier and SHAP-based feature attribution to predict Concealing, Revealing, and Faking. Binary deception prediction reaches up to 74% accuracy, while a 3-class task yields approximately 46–49% accuracy, with saccade-related metrics and maximum pupil size consistently highlighted as strong predictors. The findings demonstrate the feasibility of gaze-enabled AI for deception detection and suggest future work in higher-stakes settings and multimodal integration to improve robustness and applicability.

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 where 87 participants revealed, concealed, or faked the value of a previously selected card. The second, collected with Pupil Neon, involved 36 participants performing a similar task but facing an experimenter. XGBoost achieved accuracies 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, amplitude, and 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.

Eye Movements as Indicators of Deception: A Machine Learning Approach

TL;DR

This study investigates whether eye movements can serve as indicators of deception within Concealed Information Tests (CIT). It evaluates gaze and pupil features from two datasets—Eyelink 1000 (computer-based) and Pupil Neon (wearable)—using an XGBoost classifier and SHAP-based feature attribution to predict Concealing, Revealing, and Faking. Binary deception prediction reaches up to 74% accuracy, while a 3-class task yields approximately 46–49% accuracy, with saccade-related metrics and maximum pupil size consistently highlighted as strong predictors. The findings demonstrate the feasibility of gaze-enabled AI for deception detection and suggest future work in higher-stakes settings and multimodal integration to improve robustness and applicability.

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 where 87 participants revealed, concealed, or faked the value of a previously selected card. The second, collected with Pupil Neon, involved 36 participants performing a similar task but facing an experimenter. XGBoost achieved accuracies 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, amplitude, and 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.
Paper Structure (23 sections, 2 figures, 1 table)

This paper contains 23 sections, 2 figures, 1 table.

Table of Contents

  1. Introduction
  2. Related Works
  3. Present Study
  4. Experimental Methods
  5. Design
  6. Eyelink Dataset.
  7. Neon Dataset.
  8. Apparatus & Stimuli
  9. Participants
  10. Preparation of the Data
  11. Prediction of Deceptive Intentions
  12. Datasets
  13. Target Distribution.
  14. Feature Datasets.
  15. Model and Training Methodology
  16. ...and 8 more sections

Figures (2)

  • Figure 1: Experiments from both datasets contain the selection of a target, followed by the presentation of a baseline card and the number cards one after another for 5s each. In the Eyelink dataset, the number cards are separated by baseline cards for 2s. Participants are instructed to use their eyes to reveal, conceal, or fake the value of their target to either a computer or an experimenter.
  • Figure :