Schizophrenia Detection Using Eye Movement Data: A Machine Learning–Based Comparative Study

Abstract

Schizophrenia is a severe mental disorder associated with impairments in perception, cognition, and visual exploration behavior. Objective and non-invasive screening approaches based on eye movement analysis have emerged as promising tools to complement traditional clinical assessment. This study investigates the feasibility of detecting schizophrenia using eye tracking data through a comparative machine learning framework. Two complementary feature-engineering strategies are developed and evaluated at the subject level. The first strategy focuses on a compact set of clinically interpretable biomarkers derived from fixation dynamics, saccadic characteristics, pupil variability, and spatial distribution complexity. The second strategy constructs a high-dimensional behavioral representation by modeling gaze transitions, scanpath topology, and oculomotor statistics, followed by a structured feature selection procedure to improve generalization. Both pipelines are assessed using stratified cross-validation on a training cohort and final evaluation on an independent held-out test set. The experimental results on a benchmark dataset demonstrate that eye movement patterns contain discriminative information sufficient for automated schizophrenia recognition. Our two approaches achieved highest accuracy of 85.42\% and 81.25\% respectively. While the high-dimensional representation benefits from non-linear modeling after feature selection, the low-dimensional biomarker approach achieves competitive performance with reduced complexity and stronger interpretability. These findings highlight the trade-off between representational richness and clinical transparency and suggest that feature-centric machine learning models can provide practical and scalable support for data-driven mental health screening. The study contributes methodological insights that may facilitate translational development of interpretable, eye movement–based decision support systems in psychiatric assessment.