Classification of Mental Arithmetic Based Hybrid EEG+NIRS Signals

Brain-computer interface (BCI) is a communication system between brain and computer. Although the results of BCI studies are relatively successful, it is still an area that needs to be improved. Recent studies show that combining multiple signal recording methods (hybrid), which compensates for each other's disadvantages, will improve the performance of the BBA system. Electroencephalography + near infrared spectroscopy (EEG+NIRS) based systems have gained importance among hybrid BCI models in recent years. In this study, it was aimed to improve the system performance by working with two-class mental arithmetic based EEG+NIRS dataset which was recorded from 29 subjects. EEG oxygenated hemoglobin and deoxygenated hemoglobin signals were extracted Higuchi fractal dimension or autoregressive method based features. The extracted features were classified by k-nearest neighborhood, linear discrimination analysis (LDA), naive Bayes, decision tree, support vector machines and random forest methods. The best classification accuracy was calculated as 94.08% on average for the hybrid model with LDA. 9.31% better result was achieved with the hybrid model compared to EEG shows that the proposed method is effective for this data set.