Machine Learning-Based Classification of SSVEP at 40Hz Induced by Diverse Optical Parameters

Steady-state visual evoked potentials (SSVEP) have been extensively employed in brain-computer interfaces (BCI) using frequency modulation. While low-frequency SSVEP boasts a high signal-to-noise ratio (SNR) and robust stability, it is associated with poor visual comfort. Conversely, high-frequency SSVEP offers enhanced visual comfort but suffers from a lower SNR. In this study, we record and analyze 50 instances of high-frequency SSVEP at 4 0 H z, varying optical parameters such as illuminance, color temperature, and light source area. To achieve a precise classification of the electroencephalogram (EEG) signal, we test nine machine learning (ML) classifiers. Notably, the extreme random tree (ERT) classifier demonstrated an impressive classification accuracy of 9 2. 8 3%. Additionally, we employee SHapley Additive exPlanation (SHAP) for interpretability analysis of the ML results. Our findings indicate that the β band of the power spectral density (PSD) predominantly contributes to the classification accuracy. These results reveal the SSVEP response characteristics of different light parameters at 40 Hz. The interpretability analysis helps to build a more efficient stimulation paradigm and parameter selection mechanism, providing theoretical support for the individualization and high robustness of the BCI system.