Eye movement evidence in investigative identification based on experiments

Abstract

Investigative identification is a routine criminal investigative procedure, the results of which can be used as evidence in litigation. However, some suspects often deny their involvement in the case, and some witnesses may withhold information or misrepresent it, all of which may lead to a miscarriage of justice. This study created a stressful environment and conducted a simulated crime experiment to explore whether eye movement data can be an effective feature for distinguishing perpetrators, innocents, and insiders. The eye movement features—such as the total fixation duration, number of fixations, and first fixation duration—within an area of interest were collected from 83 participants sorted into informed, involved, and innocent groups. The results revealed the following: (1) compared with the object and scene stimuli, subjects with different identities were more likely to exhibit significant differences in eye movement data for the involved and irrelevant portraits. The total fixation duration and the number of fixations can provide a reference for judging whether someone is involved in a case, and the first fixation duration effect was not obvious. (2) Using machine learning algorithms to predict subjects’ identities through eye movement features, it was demonstrated that the involved portrait-object-scene model had the best predictive effect. (3) Multiple algorithmic models were used to distinguish subjects’ identities, and the highest accuracy of 92.7% was achieved for the informed × innocent group, 88% for the innocent × suspect group (including the informed and involved groups), and 84.5% for the involved group. The eye movement analysis method can provide a reference for criminal investigators to distinguish between the perpetrator, insider, and innocent, and offer a novel approach to determining the direction of further investigation and uncovering and verifying case clues.