Quaternion Interpolation for Augmenting Eye Dynamic Signals: Tackling Class Imbalance in Depression Severity Prediction Using Smart Eyewear

Accurate prediction of depression severity is critical for enabling timely and effective interventions. Eye dynamics have recently emerged as a promising biomarker for assessing mental health conditions, including depression. However, the inherent class imbalance in depression datasets poses significant challenges to the development of robust predictive models. In this study, a novel approach is proposed to address this issue by augmenting eye movement signals using quaternion interpolation. Quaternion interpolation enables the generation of realistic and continuous eye gaze trajectories, thereby enriching the dataset and mitigating the effects of class imbalance. The effectiveness of this approach was evaluated on a benchmark depression dataset, demonstrating an 8.53% improvement in root mean square error and a 10.7% improvement in mean absolute error for severity prediction compared to traditional data augmentation methods. This study highlights the potential of quaternion interpolation as a powerful tool for data augmentation in mental health diagnostic, paving the way for more reliable and accurate assessments of depression severity.