Enhancing Typing Speed of P300 Speller Using Ensemble of Support Vector Machines with Dataset Manipulation for Increased Diversity

Özet

The P300 speller is a valuable brain-computer interface system that allows typing by analyzing electroencephalogram (EEG) signals generated in response to visual stimuli. Among the classification methods employed for the P300 speller, the ensemble of support vector machines (eSVM) is renowned for achieving high accuracy. However, existing eSVM approaches primarily focus on individual classifier accuracy, overlooking the importance of diversity among the classifiers. To address this limitation, we propose a dataset manipulation method that divides the training dataset into distinct groups based on the characteristics of EEG signals generated at different distances between the target letter and the visual keyboard. By training each individual SVM classifier on these diverse groups, we enhance the ensemble's diversity. Experimental results demonstrate that the proposed eSVM approach with increased diversity significantly improves the typing speed of the P300 speller, achieving an average accuracy of 70% with only four repetitions per letter, enabling verbal communication via the Language Support Program.