Performance of Extreme Learning Machine Kernels in Classifying EEG Signal Pattern of Dyslexic Children in Writing

Abstract

Dyslexia is a specific learning disability that causes leaners to have difficulties to process letters and number during reading, writing and doing mathematics. Early identification of dyslexic characteristic is crucial so that early intervention given could overcome learner difficulties. A process of writing involves areas in brain learning pathway and motor cortex. This activity could be recorded using electroencephalogram (EEG) non-invasively. Using this information, a study has been conducted to distinguish EEG signal of normal, poor and capable dyslexic children. In this work, EEG signals were recorded from eight channels; C3, C4, P3, P4, FC5, FC6, T7 and T8. The signals were extracted using discrete wavelet transform (DWT) with Daubechies wavelet family order 2, 4, 6 and 8 to acquire beta and theta band features. The coefficient of beta band power and the ratio of theta/beta band power were input features of expert learning machine (ELM) classifier. Four types of kernels namely linear, radial basis function (RBF), polynomial and wavelet were applied as output weight in connecting hidden node and the output node of ELM. Parameters were varied to optimize each kernel to obtain the best classification accuracy. Results show that db2 gives the highest classification performance for all kernel among other Daubechies family. RBF and wavelet kernel yield the highest accuracy at 89% compared with other ELM kernels. This work reveals that ELM with RBF and wavelet kernel together with beta band power and ratio of theta/beta band power extracted from db2 could distinguish normal, poor and dyslexic children during writing.