Processed Eye-Tracking Datasets for HCDG-Net Cross-Dataset Depression Assessment

The files contain subject and trial identifiers, trial order, event order, timestamps, normalized gaze coordinates, gaze displacement, fixation duration, relative time, stimulus category, functional task role, transition state, trial-level valid-event ratio, data-split information, and task-specific target variables.</p><p dir="ltr">To provide a common representation across heterogeneous affective paradigms, stimulus categories are mapped to functional task roles.

Sad stimuli are mapped to the Negative role, while Happy stimuli in Dataset A and Neutral stimuli in Dataset B are mapped to the Comparator role. Gaze transitions are represented using Stay, Leave, and Return states.</p><p dir="ltr">Spatial coordinates are normalized according to screen dimensions. Event time is normalized relative to the first and last valid events within each trial.

Fixation duration is normalized by the mean valid-event duration within the trial followed by a log1p transformation. Subject identity, trial order, and within-trial event order are preserved throughout preprocessing.</p><p dir="ltr">Both datasets include 10 subject-level repeated train/validation/test partitions. For each repetition, 210 subjects are assigned to training, 45 to validation, and 45 to testing.

The split columns allow the experimental protocol reported in the associated HCDG-Net study to be reproduced.</p><p dir="ltr">Dataset A file: HCDG_Dataset_A.csv<br>Dataset B file: HCDG_Dataset_B.csv</p><p dir="ltr">The dataset is intended for research on eye-tracking analysis, depression assessment, gaze dynamics modeling, cross-dataset learning, domain adaptation, and heterogeneous-supervision learning.</p>