A multiple linear regression model approach for two-class fNIR data classification

Abstract

Listen

Translate article

Functional near-infrared (fNIR)-based motor imagery (MI) classification is an interesting challenge for the brain–computer interface (BCI) implementation. The success of the fNIR-BCI mostly depends on the classification performance. Usually, machine learning-based complex algorithms are used to classify the fNIR-MI data which often proves impractical to program in small hardware. Therefore, the practical small and low price hardware-based BCI system demands simple algorithm that can provide satisfactory classification accuracy. This paper proposes an fNIR-MI data classifier utilizing the multiple linear regression model (MLRM). In this work, two-class fNIR-MI data were acquired for feature extraction and classification. The predictive model for classification was prepared by the proposed method. Additionally, the same dataset was classified by the widely used classification method—support vector machine (SVM) and linear discriminant analysis (LDA). The results reveal that the proposed multiple linear regression model-based classifier (MLRMC) shows almost equivalent results compared to the SVM and LDA on the same features, although MLRM is a much simpler algorithm than that of the others. Therefore, the outcomes of this research work claim that the proposed MLRMC is a good choice as a classifier for fNIR-like simple signal classification.