A Data-Driven Unsupervised Framework for Discovering Interpretable Gaze-Based Behavioral Pseudo-Zones in Children...

Background/Objectives: Children with autism spectrum disorder (ASD) often exhibit differences in attention regulation and visual behavior. However, many ASD eye-tracking datasets lack reliable moment-to-moment behavioral or emotional annotations, limiting the direct application of supervised learning approaches. To address this challenge, this study proposes an interpretable gaze-based unsupervised framework for discovering behavioral pseudo-zones from unlabeled ASD eye-tracking data. Methods: Raw gaze recordings from ASD participants were segmented into fixed temporal windows and represented using interpretable gaze features, including gaze dispersion, fixation duration, tracking quality, motion ratio, pupil size, and gaze velocity measures. Multiple clustering models and alternative temporal window sizes were systematically compared, including K-means, Gaussian Mixture Modeling (GMM), Agglomerative Clustering, and HDBSCAN. Results: Among the evaluated configurations, the combination of 1000 ms windows with K-means clustering (k = 4) was retained as the final exploratory configuration. Although alternative solutions achieved slightly stronger internal validation metrics, the selected configuration provided a more interpretable four-zone structure while maintaining acceptable clustering quality. The final retained solution produced four interpretable behavioral pseudo-zones with statistically significant differences across all extracted gaze features according to the Kruskal-Wallis test (p < 0.05). A PCA projection further supported the exploratory structure of the discovered pseudo-zones, with the first two principal components explaining 72.3% of the total variance. Conclusions: The findings demonstrate that unlabeled ASD gaze data can be organized into interpretable behavioral pseudo-zones using an unsupervised and transparent feature-based framework. This work contributes a data-driven and interpretable framework for future gaze-based behavioral analysis and autism-related AI research.