Comparison Between Support Vector Machine with Polynomial and RBF Kernels Performance in Recognizing EEG Signals of Dyslexic Children

Abstract

Dyslexia is seen as learning disorder that causes learners having difficulties to recognize the word, be fluent in reading and to write accurately. This is characterized by a deficit in the region associated with learning pathways in the brain. Activities in this region can be investigated using electroencephalogram (EEG). In this work, Discrete Wavelet Transform (DWT) with Daubechies order of 2 (db2) based features extraction was applied to the EEG signal and the power is calculated. The differences between beta and theta band with responding to learning activities were explored. Multiclass Support Vector Machine (SVM) was used to classify the EEG signal. Performance comparison of Polynomial and Radial Basis Function (RBF) kernel recognizing EEG signal during writing word and non-word is presented in this paper. It was found that SVM with RBF kernel performance was generally higher than that of the polynomial kernel in recognizing normal, poor and capable dyslexic children. The SVM with RBF kernel produced 91% accuracy compared to the polynomial kernel.