Extreme Learning Machine for Distinction of EEG Signal Pattern of Dyslexic Children in Writing

Abstract

Dyslexia is neurological disorder that affects the brain ability to process symbols such as letters and numbers. The process of writing involves learning pathway that can be monitored non-invasively using electroencephalogram (EEG). Majority EEG based studies on dyslexia have been on reading. Here, in this paper, an extreme learning machine (ELM) classifier with radial basis function (RBF) kernel is employed to distinguish between normal, poor and capable dyslexic subjects, from EEG signals of their writing. The RBF kernel allows its center and width randomly to be generated, such that the output weights of RBF networks can be calculated analytically instead of being iteratively tuned, resulting in faster learning speed. Power band coefficients of beta and beta/theta ratio are extracted using discrete wavelet transform (DWT) with Daubechies family order 2, 4, 6 and 8 to serve as inputs to the classifier. From the experimental results, it is found that Db2 yields the highest accuracy at 89% and the best ROC performance for the three cohorts.