Home // eKNOW 2016, The Eighth International Conference on Information, Process, and Knowledge Management // View article

fNIRS Neural Signal Classification of Four Finger Tasks using Ensemble Multitree Genetic Programming

Authors:
Jinung An
Jong-Hyun Lee
Sang Hyeon Jin
Chang Wook Ahn

Keywords: Classification; Finger Tasks; Neural Signal; Emsemble Learning; Mutitree Genetic Programming

Abstract:
Accuracy of classification and recognition in neural signal is the most important issue to evaluate the clinical assessment or extraction of features in brain computer interface. Especially, classification of multitasks by measuring functional Near-Infrared Spectroscopy (fNIRS) is a challenging due to its low spatiotemporal resolution. To improve the classification accuracy of fNIRS neural signals for multitasks, an evolutionary computing method was proposed. Four healthy participants performed four finger tasks which are digit-active, digit-passive, thumb-active and thumb-passive. To classify the four tasks, a multitask classifier was devised by the ensemble multitree genetic programming (EMGP). The experimental results validate the performance of the proposed classifier. The comparison of the conventional and proposed classifiers at the real classification experiment shows the higher accuracy of the proposed method. Moreover, it reveals the improvement of classification accuracy when compared with conventional classifiers in the additional experiment of fifteen dataset in University of California Irvine machine learning repository. The proposed classifier can be effective to classify and recognize the fNIRS neural signals during multitasks. Moreover, the subject dependent learning can be designed for the local brain activation training based on neuro-feedback. After data learning for all classes, the subject tries to make their brain activation of an active task as similar with a passive task by the online motor-imagery with action observation. As a result, the subject is trained to concentrate his brain activation for the essential area of brain. The proposed classifier can be applied well because high classification accuracy is essential to the neuro-training system. Finally, the classification accuracy of the proposed EMGP is 5.48% higher than the average of conventional classifiers.

Pages: 14 to 18

Copyright: Copyright (c) IARIA, 2016

Publication date: April 24, 2016

Published in: conference

ISSN: 2308-4375

ISBN: 978-1-61208-472-5

Location: Venice, Italy

Dates: from April 24, 2016 to April 28, 2016