Improving the performance of brain computer interface system using electroencephalography and near infrared spectroscopy-based hybrid model

Brain Computer Interface (BCI)'s can use signals from various brain activity measuring devices as inputs. Among these, electroencephalography (EEG) is widely used in BCI studies because it has some advantages over other methods. However, the ineffectiveness of the performance of BCI systems with these studies, the prevalence of studies with similar experimental content, and the fact that studies with new experimental content have some disadvantages lead researchers to experimental paradigm-based studies. In addition, neuroimaging methods for recording neural activity have their own advantages and disadvantages. Based on this point of view, it can be said that combining multiple signal recording methods that compensate for each other's disadvantages will improve the performance of the BCI system. Unlike EEG, among other methods, near infrared spectroscopy (NIRS) has the advantage of relative robustness against body movements and electrical products. In this thesis, a 4-class hybrid EEG+ NIRS dataset with a unique experimental paradigm was recorded. A high performance BCI system is proposed by using hybrid modality of this original dataset based on right, left, up and down scrolling text reading. The EEG+ NIRS dataset was recorded at Atatürk University with the support of the Karadeniz Technical University BAP project numbered FHD-2020-9166 and service procurement. In this thesis study, which aims to create a faster and more accurate BCI system, features are extracted from the dataset using Hilbert Transform, then the features are classified with the k-nearest neighbor algorithm and a high-performance hybrid BCI system is proposed by calculating 96.28%±1.30 classification accuracy with the hybrid model.