A Comparative Analysis of Classifiers for Steady-State Visual Evoked Potential-Based BCIs

Steady-state visual Evoked Potentials (SSVEPs) are widely used in Brain-Computer Interfaces (BCIs), but accurate frequency classification remains challenging due to inter-subject variability and complex signal patterns. This study aims to improve SSVEP frequency prediction by combining wavelet decomposition and testing it with several classification models - Support Vector Machines (SVM), k-nearest Neighbors (kNN), Decision Trees (DT), and Random Forest (RF). Using EEG data from 15 healthy subjects, we observe that accuracy is higher with fewer classes but decreases as the number of classes increases. Notably, SVM outperforms other models significantly, achieving an average accuracy of 92.5 % in higher-class configurations of 40 frequencies, while <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\text{kNN}, \text{DT}$</tex>, and RF show substantially lower performance in similar setups. For 40-class configurations, SVM's accuracy increases from 41.3 % with one block of training data to 92.5 % with five blocks, underscoring that training data size boosts performance by over <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$5 0 \%$</tex>. Our findings provide valuable insights for developing more efficient and reliable BCI systems.