Efficient Mental Arithmetic Task Classification using Wavelet Domain Statistical Features and SVM Classifier

Abstract

Functional Near Infrared Spectroscopy (fNIRS) has

been emerged as a potential technique in the research of

BCI. In this paper, we proposed a discrete wavelet transform

based feature extraction technique to classify mental arithmetic

tasks from fNIRS data. In order to investigate the change in

brain activities during mental arithmetic task, recorded data

are windowed in several frames. DWT has been employed on

different channels of each frame and then a number of statistical

features are extracted from both the approximate and the detail

coefficients of data in order to distinguish the mental arithmetic

task and the rest condition. Six-fold cross validation is performed

using SVM classifier to examine the effectiveness of DWT based

features. Efficacy of oxyhemoglobin, deoxyhemoglobin, and total

hemoglobin data from different selected channel combinations

are also examined. It is observed that proposed algorithm

provides a satisfactory accuracy of 93.26% using DWT based

features extracted from 104 channels.