Gaze Prediction as a Function of Eye Movement Type and Individual Differences

TL;DR

This study investigates gaze-prediction performance across eye-movement types and individuals using three distinct models: a lightweight LSTM, a transformer-based time-series model, and an anatomically inspired OPKF with Kalman filtering. Evaluated on the Gazebase dataset (Round 1, 322 participants) with a short prediction interval, the work reveals robust subject-to-subject variability and clear performance ordering across eye-movement types, with fixation yielding the best, and large saccades the worst. The authors identify oculomotor indices—such as fixation-noise thresholds and saccade-velocity metrics—that consistently predict prediction difficulty, and they demonstrate that inter-subject variation is substantial enough to warrant reporting and model-tuning to reduce dispersion. The findings have practical implications for gaze-based XR rendering and HCI, highlighting the need to incorporate individual-difference statistics and per-subject cues to improve real-time gaze prediction robustness.

Abstract

Eye movement prediction is a promising area of research with the potential to improve performance and the user experience of systems based on eye-tracking technology. In this study, we analyze individual differences in gaze prediction performance. We use three fundamentally different models within the analysis: the lightweight Long Short-Term Memory network (LSTM), the transformer-based network for multivariate time series representation learning (TST), and the Oculomotor Plant Mathematical Model wrapped in the Kalman Filter framework (OPKF). Each solution was assessed on different eye-movement types. We show important subject-to-subject variation for all models and eye-movement types. We found that fixation noise is associated with poorer gaze prediction in fixation. For saccades, higher velocities are associated with poorer gaze prediction performance. We think these individual differences are important and propose that future research should report statistics related to inter-subject variation. We also propose that future models should be designed to reduce subject-to-subject variation.

Figures (10)

• Figure 1: CDF Plots Across All Eye-Movement Events
• Figure 1: CDF Plots Across All Eye-Movement Events and Fixations Only
• Figure 2: Gaze Prediction as Function of Eye-Movement Type
• Figure 2: Gaze Prediction Error as a Function of Saccade Size Across PIs
• Figure 3: Subject Profiles in Gaze Prediction Error