A hierarchical meta-model for multi-class mental task based brain-computer interfaces

Akshansh Gupta,R.K Agrawal,Jyoti Singh Kirar,Baljeet Kaur,Weiping Ding,Chin-Teng Lin,Javier Andreu-Perez,Mukesh Prasad

doi:10.1016/j.neucom.2018.07.094

Abstract

In the last few years, many research works have been suggested on Brain-Computer Interface (BCI), which assists severely physically disabled persons to communicate directly with the help of electroencephalogram (EEG) signal, generated by the thought process of the brain. Thought generation inside the brain is a dynamic process, and plenty thoughts occur within a small time window. Thus, there is a need for a BCI device that can distinguish these various ideas simultaneously. In this research work, our previous binary-class mental task classification has been extended to the multi-class mental task problem. The present work proposed a novel feature construction scheme for multi mental task classification. In the proposed method, features are extracted in two phases. In the first step, the wavelet transform is used to decompose EEG signal. In the second phase, each feature component obtained is represented compactly using eight parameters (statistical and uncertainty measures). After that, a set of relevant and non-redundant features is selected using linear regression, a multivariate feature selection approach. Finally, optimal decision tree based support vector machine (ODT-SVM) classifier is used for multi mental task classification. The performance of the proposed method is evaluated on the publicly available dataset for 3-class, 4-class, and 5-class mental task classification. Experimental results are compared with existing methods, and it is observed that the proposed plan provides better classification accuracy in comparison to the existing methods for 3-class, 4-class, and 5-class mental task classification. The efficacy of the proposed method encourages that the proposed method may be helpful in developing BCI devices for multi-class classification.

Highlights

The human brain has the capability of differentiating multiple courses of action without any difficulty
This motivated us to investigate feature selection method for multi-mental task classification problem. It has been observed in the research work [10] that the combination of feature extraction using Wavelet transform (WT) and feature selection using Linear Regression (LR) has given the best set of features that enhance the performance of the classifier for the binary mental task classification
Features are extracted from the EEG signal in two steps: (1) EEG signal is decomposed by Wavelet Transform (WT) and (2) phase statistical, and uncertainty parameters are calculated from each decomposed signal to represent the signal more compactly

Summary

Introduction

The human brain has the capability of differentiating multiple courses of action without any difficulty. To the best of our knowledge, feature selection has not been suggested in research work related to multi-class mental task classification. This motivated us to investigate feature selection method for multi-mental task classification problem. It has been observed in the research work [10] that the combination of feature extraction using Wavelet transform (WT) and feature selection using Linear Regression (LR) has given the best set of features that enhance the performance of the classifier for the binary mental task classification. The proposed method utilized Optimal decision tree based on amalgamation with support vector machine (SVM) to build decision model to distinguish multiple mental tasks.

Related Works

Feature Extraction

Feature Selection

Experimental Set-up and Result

Conclusion