Statistical Evaluation of Factors Influencing Inter-Session and Inter-Subject Variability in EEG-Based Brain Computer Interface

A cognitive alteration in the form of diverse mental states has a significant impact on performance electroencephalography (EEG) based brain computer interface (BCI). Such alterations include change concentration levels commonly recognized as being indicated by alpha rhythm, drowsiness or fatigue which occurs during EEG signal acquisition. Change state give rise to challenge variability characteristics across sessions and subjects. Consequently, this constitutes low intention detection rate (IDR) that renders BCI unreliable. This study investigates multiple factors lead poor EEG-BCI. Five (i) level, (ii) selection independent components(IC), (iii) inter-session variability, (iv) inter-subject (v) classification methods IDR BCI. The indicator is validated, relationship between rhythm studied among sessions. In addition, ICs are examined determine their effects possibility two contain similar also examined, where both acquired from same subject different days. Moreover, subjects containing examined. Furthermore, conquer dynamics feature transfer learning (TL) approach proposed study. three (TL, KNN NB) compared whether multi-source neural information can improve accuracy individual Three datasets using paradigms used for experiments. steady motion visual evoked potential (SSMVEP), motor imagery (MI) competition IV-a dataset. Experimental results have shown components an effect IDR. case IC-2 IC-11 achieved lowest highest accuracies 51% 100% SSMVEP datasets, while IC-9 double-component (IC-2 IC-13) 40% 69% MI respectively. second experiment demonstrated higher depicted lower corresponds level. While within significantly deteriorate As such decline 82% 61%, 56% 44% was observed Integration samples but resulted 65%, 59% SSMVEP, 48% datasets. When classifiers evaluated domains, NB 52% respectively, TL showed increase with 98% similarly manner 49% 42% respectively subjects, 64% achieved. when 9 dataset, 68% 65% 99% conclusion, magnitude affect component due non-linear non-stationary nature signals. merging sessions, factor introduce challenges overfitting resulting found critical, because some advanced accuracy.