Improving P300 Speller performance by means of optimization and machine learning

Abstract Brain-Computer Interfaces (BCIs) are systems allowing people to interact with the environment bypassing natural neuromuscular and hormonal outputs of peripheral nervous system (PNS). These interfaces record a user’s brain activity translate it into control commands for external devices, thus providing PNS additional artificial outputs. In this framework, BCIs based on P300 Event-Related Potentials (ERP), which represent electrical responses recorded from after specific events or stimuli, have proven be particularly successful robust. The presence absence evoked potential within EEG features is determined through classification algorithm. Linear classifiers such as stepwise linear discriminant analysis support vector machine (SVM) most used algorithms ERPs’ classification. Due low signal-to-noise ratio signals, multiple stimulation sequences (a.k.a. iterations) carried out then averaged before signals being classified. However, while augmenting number iterations improves Signal-to-Noise Ratio, also slows down process. early studies, was fixed (no stopping environment), but recently several strategies been proposed in literature dynamically interrupt sequence when certain criterion met order enhance communication rate. work, we explore how improve performances by combining optimization learning. First, propose new decision function that aims at improving terms accuracy Information Transfer Rate both no environment. Then, SVM training problem facilitate target-detection Our approach proves effective publicly available datasets.