Support vector machine with theta-beta band power features generated from writing of dyslexic children

The classification of dyslexia using EEG requires the detection of subtle differences between groups of children in an environment that are known to be noisy and full of artifacts. It is thus necessary for the feature extraction to improve the classification. The normal and poor dyslexic are found t...

Full description

Bibliographic Details
Published in:International Journal of Integrated Engineering
Main Author: Mahmoodin Z.; Lee K.Y.; Mansor W.; Zainuddin A.Z.A.
Format: Article
Language:English
Published: Penerbit UTHM 2019
Online Access:https://www.scopus.com/inward/record.uri?eid=2-s2.0-85075880383&doi=10.30880%2fijie.2019.11.03.005&partnerID=40&md5=d3e80edae76546f5c4a7bce69c9e5e7e
id 2-s2.0-85075880383
spelling 2-s2.0-85075880383
Mahmoodin Z.; Lee K.Y.; Mansor W.; Zainuddin A.Z.A.
Support vector machine with theta-beta band power features generated from writing of dyslexic children
2019
International Journal of Integrated Engineering
11
3
10.30880/ijie.2019.11.03.005
https://www.scopus.com/inward/record.uri?eid=2-s2.0-85075880383&doi=10.30880%2fijie.2019.11.03.005&partnerID=40&md5=d3e80edae76546f5c4a7bce69c9e5e7e
The classification of dyslexia using EEG requires the detection of subtle differences between groups of children in an environment that are known to be noisy and full of artifacts. It is thus necessary for the feature extraction to improve the classification. The normal and poor dyslexic are found to activate similar areas on the left hemisphere during reading and writing. With only a single feature vector of beta activation, it is difficult to distinguish the difference between the two groups. Our work here aims to examine the classification performance of normal, poor and capable dyslexic with theta-beta band power ratio as an alternative feature vector. EEG signals were recorded from 33 subjects (11 normal, 11 poor and 11 capable dyslexics) during tasks of reading and writing words and non-words. 8 electrode locations (C3, C4, FC5, FC6, P3, P4, T7, T8) on the learning pathway and hypothesized compensatory pathway in capable dyslexic were applied. Theta and beta band power features were extracted using Daubechies, Symlets and Coiflets mother wavelet function with different orders. These are then served as inputs to linear and RBF kernel SVM classifier, where performance is measured by Area Under Curve (AUC) of Receiver Operating Characteristic (ROC) graph. Result shows the highest average AUC is 0.8668 for linear SVM with features extracted from Symlets of order 2, while 0.9838 for RBF kernel SVM with features extracted from Daubechies of order 6. From boxplot, the normal subjects are found to have a lower theta-beta ratio of 2.5:1, as compared to that of poor and capable dyslexic, ranging between 3 to 5, for all the electrodes. © Universiti Tun Hussein Onn Malaysia Publisher's Office.
Penerbit UTHM
2229838X
English
Article
All Open Access; Bronze Open Access
author Mahmoodin Z.; Lee K.Y.; Mansor W.; Zainuddin A.Z.A.
spellingShingle Mahmoodin Z.; Lee K.Y.; Mansor W.; Zainuddin A.Z.A.
Support vector machine with theta-beta band power features generated from writing of dyslexic children
author_facet Mahmoodin Z.; Lee K.Y.; Mansor W.; Zainuddin A.Z.A.
author_sort Mahmoodin Z.; Lee K.Y.; Mansor W.; Zainuddin A.Z.A.
title Support vector machine with theta-beta band power features generated from writing of dyslexic children
title_short Support vector machine with theta-beta band power features generated from writing of dyslexic children
title_full Support vector machine with theta-beta band power features generated from writing of dyslexic children
title_fullStr Support vector machine with theta-beta band power features generated from writing of dyslexic children
title_full_unstemmed Support vector machine with theta-beta band power features generated from writing of dyslexic children
title_sort Support vector machine with theta-beta band power features generated from writing of dyslexic children
publishDate 2019
container_title International Journal of Integrated Engineering
container_volume 11
container_issue 3
doi_str_mv 10.30880/ijie.2019.11.03.005
url https://www.scopus.com/inward/record.uri?eid=2-s2.0-85075880383&doi=10.30880%2fijie.2019.11.03.005&partnerID=40&md5=d3e80edae76546f5c4a7bce69c9e5e7e
description The classification of dyslexia using EEG requires the detection of subtle differences between groups of children in an environment that are known to be noisy and full of artifacts. It is thus necessary for the feature extraction to improve the classification. The normal and poor dyslexic are found to activate similar areas on the left hemisphere during reading and writing. With only a single feature vector of beta activation, it is difficult to distinguish the difference between the two groups. Our work here aims to examine the classification performance of normal, poor and capable dyslexic with theta-beta band power ratio as an alternative feature vector. EEG signals were recorded from 33 subjects (11 normal, 11 poor and 11 capable dyslexics) during tasks of reading and writing words and non-words. 8 electrode locations (C3, C4, FC5, FC6, P3, P4, T7, T8) on the learning pathway and hypothesized compensatory pathway in capable dyslexic were applied. Theta and beta band power features were extracted using Daubechies, Symlets and Coiflets mother wavelet function with different orders. These are then served as inputs to linear and RBF kernel SVM classifier, where performance is measured by Area Under Curve (AUC) of Receiver Operating Characteristic (ROC) graph. Result shows the highest average AUC is 0.8668 for linear SVM with features extracted from Symlets of order 2, while 0.9838 for RBF kernel SVM with features extracted from Daubechies of order 6. From boxplot, the normal subjects are found to have a lower theta-beta ratio of 2.5:1, as compared to that of poor and capable dyslexic, ranging between 3 to 5, for all the electrodes. © Universiti Tun Hussein Onn Malaysia Publisher's Office.
publisher Penerbit UTHM
issn 2229838X
language English
format Article
accesstype All Open Access; Bronze Open Access
record_format scopus
collection Scopus
_version_ 1812871800232607744