Effect of Pre-Processing in Four Class of Functional Near-Infrared Spectroscopy for Brain-Computer Interface

Brain-computer interface (BCI) is one of the technologies that help people with impairment. Functional Near-Infrared Spectroscopy (fNIRS) is a non-invasive tool frequently used for BCI applications. Data arrangement of fNIRS-BCI and feature extraction have crucial roles in the accuracy improvement of machine learning algorithms in BCI applications. Furthermore, Partial Least Square Regression (PLS) and Principal Component Analysis feature popular extractions in medical fields such as NIRS. However, the fNIRS data barely finds the feature extraction and data arrangement combinations. Therefore, this paper compares the analysis of classification results of two different data arrangement data with a combination of PLS regression and PCA. The data arrangement is a combination of HbO and HbR, and HbO with a statistical approach, which is mean, a combination of mean with skewness, kurtosis, standard deviation, variance, skewness, and slope with peak. However, specific components are applied to the feature extraction into four PCs. The PCs were used for performance study by using KNN with adjusted K-Neighborhood. The results are being compared to find a suitable data arrangement for fNIRS-BCI. The finding shows that data arrangement with fewer PCs outperformed that with more PCs.