Support Vector Machine Classification of EEG Signals for Word and Non-Word Writing in Normal, Poor Dyslexic and Capable Dyslexic Children

摘要

Electroencephalogram (EEG) signal provides an insight into the workings of the brain by recording of electrical activity on the surface of the scalp. The differences in the EEG signal neural representations from the norm indicates abnormality and thus classifiable. It is known thatdyslexia, a neurological disorder that impairs the subject ability to properly read or write, is caused by the inefficiency of the brain’s left hemisphere processing region. This study looks into the classification of EEG signals of writing word, non-word and a combination of both inidentifying subjects as either normal, poor dyslexic or as capable dyslexic. With an overall total of 21 subjects having an age range of 6 to 11 years old band power features were extracted from EEG signals recorded from eight (8) electrode locations which are, C3, C4, P3, P4, T7, T8, FC5and FC6 that corresponds to the brain’s learning pathway using Daubechies wavelet transform of order 2. A multiclass support vector machine (SVM) was applied as the classification algorithm with the classifier having three (3) different training data set of word, non-word and combinationof both. Results showed an improved classifier performance with the combine training sets of word and non-word with an accuracy of 100% for word test data and 83.3% for non-word. Applying only word for the training data yielded an accuracy of 75%, and for the non-word, it is 58%.