EEG BASED MENTAL WORKLOAD CLASSIFICATION USING HILBERT HUANG TRANSFORM

Abstract

This work aims to classify human stress levels for rest v/s mental arithmetic task. In this work, a publicly available EEG During Mental Arithmetic Tasks dataset is used, comprising of EEG data of 36 participants. A 23-channel EEG device was used for collecting the data while the participants were solving arithmetic problems. This induces a short-time stress which is captured by the EEG device. For efficient classification of stress levels pre-processing is done by applying filters and Independent Component Analysis. This study employs the Hilbert Huang Transform for determining the Time-Frequency aspect of feature extraction which was not considered in prior studies utilising this dataset. Features namely variance, mean frequency, maximum frequency and sample entropy are computed on the dataset. We also apply the feature selection in order to determine a subset of features which contributes most to the classification accuracy of this proposed method. SVM and K-NN are used as classifiers. This work achieves a maximum accuracy of 91.6% using SVM classifier trained on the complete set of features and 100% accuracy when trained on subset of features. It is observed that accuracy of the model is significantly improved by using the feature selection method. This work highlights the efficiency of time-frequency domain features for mental workload classification.