Objective This study aimed to use eye movement characteristics to objectively analyze the psychological activities of individuals with anxiety disorders and those at high risk, facilitating early identification during self-report questionnaire responses. It provides a practical and theoretical basis for developing a new psychological screening system. Methods A total of 296 participants was divided into three groups: anxiety disorder group (98 diagnosed patients), high-risk group (99 individuals with anxiety traits and symptoms not meeting diagnostic criteria), and normal control group (99 healthy individuals). Using a multi-modal approach combining self-report questionnaires and eye-tracking, differences in questionnaire scores and basic eye movement features were analyzed. Machine learning models (Random Forest and XGBoost) were applied to classify the groups based on questionnaire scores, eye movement indicators, and their combination. Result Significant differences were found among the three groups in questionnaire scores (F = 119.165, P < 0.001). The anxiety disorder group scored significantly higher than both the high-risk and control groups (P < 0.001), and the high-risk group scored higher than the control group (P < 0.001). Eye movement indicators—total number of fixations, total fixation duration, number of regressions, and pupil diameter—also showed significant differences between clinical groups and controls (P < 0.001 or P < 0.05), though no significant difference was found between the anxiety and high-risk groups. The machine learning classification accuracy based on questionnaire scores alone was 66.2%, on eye movement alone was 49.3%, and on combined features reached 74.7%, an 8.5% improvement over using only questionnaire data. Conclusion The study revealed distinct eye movement patterns in anxiety disorder and high-risk groups during self-report questionnaire completion. Combining eye-tracking data with self-reports improved classification accuracy via machine learning, demonstrating the potential of multi-modal methods in practical settings such as psychological screening.