USING EYE MOVEMENTS TO MEASURE SKILL IN CRIME SCENE INVESTIGATION

Abstract

his thesis investigates whether eye movement patterns can be used to distinguish be-tween Control and CSI performance in the context of crime scene investigation. By ana-lyzing gaze data collected from participants of varying expertise levels, this study explores both visual behavior and classification potential using machine learning.
The research uses a dataset of eye-tracking fixations recorded during the inspection of simulated 2D crime scene images. Multiple visualization techniques (e.g., heatmaps, scan-paths, scarfplots) were employed to explore group-level differences. Additionally, gaze fea-tures such as fixation ratios and gaze entropy were extracted to quantify viewing patterns. To assess predictive capabilities, several machine learning models were tested, including Support Vector Machines (SVM), Neural Networks, and Long Short-Term Memory (LSTM) networks. While traditional models performed near or below the chance level, the LSTM achieved an accuracy of up to 87% under ideal training conditions.
Despite this, generalizability remains limited, this phenomenon was visible in the deep learning models. Differences between groups were often subtle and inconsistent across visualizations and metrics.
Overall, this thesis highlights both the potential and current limitations of using eye-tracking data to assess forensic expertise. Future research is encouraged to incorporate immersive technologies such as VR or NeRF to better simulate real-world conditions and improve classification robustness.

Date of Award	16 Aug 2025
Original language	English
Supervisor	Frouke Hermens (Examiner) & Daniel Tan (Co-assessor)