Toward Eye Tracking via Forehead-Wearable EEG: an Evaluation of Electrode Placements and Reference Schemes

Wearable EEG systems increasingly use frontal electrodes, which are highly sensitive to eye movement artifacts. While these signals are often treated as noise, recent efforts suggest they may contain useful information for estimating gaze direction and movement. However, how electrode placements and reference schemes affect eye-tracking functionality in the context of compact, wearable EEG devices with limited frontal electrode coverage remains unclear. This study provides a systematic evaluation of these factors. We simultaneously recorded EEG and eye movement data from 20 participants and evaluated the effects using seven forehead electrode pairs and four reference schemes. For each dual-channel configuration, 21 time-, frequency-, and phase-domain features were extracted. Classification and regression models were evaluated using leave-one-subject-out cross-validation. Focusing on the FP1-FP2 pair, we achieved reliable horizontal movement classification and regression, with decision tree models yielding a mean F1-score of 89.84% and Pearson correlation r=0.80 under AFZ reference. For vertical movements, the best results were obtained under A1-A2 average reference, with an F1-score of 86.16% and r=0.77. In broader comparisons, AF7-AF8 maintained robust performance across most reference schemes. Midline pairs such as AF3-AF4 showed consistently lower vertical correlations, especially under AFZ. Overall, earlobebased references provided more stable results across both directions compared to AFZ or CMS/DRL references. These findings demonstrate that eye movement artifacts contain informative signals that, when appropriately leveraged, enable lightweight EEG-based eye tracking. Even with only two frontal electrodes, such systems have demonstrated promising feasibility in gaze decoding, highlighting their potential for integration into wearable BCI, assistive communication, and context-aware interfaces.