Machine learning and deep learning performance in classifying dyslexic children’s electroencephalogram during writing

Abstract

Dyslexia is a form of learning disability that causes a child to have difficulties in writing alphabets, reading words, and doing mathematics. Early identification of dyslexia is important to provide early intervention to improve learning disabilities. This study was carried out to differentiate EEG signals of poor dyslexic, capable dyslexic, and normal children during writing using machine learning and deep learning. three machine learning algorithms were studied: k-nearest neighbors (KNN), support vector machine (SVM), and extreme learning machine (ELM) with input features from coefficients of beta and theta band power extracted using discrete wavelet transform (DWT). As for the deep learning (DL) algorithm, long short-term memory (LSTM) architecture was employed. The kernel parameters of the classifiers were optimized to achieve high classification accuracy. Results showed that db8 achieved the greatest classification accuracy for all classifiers. Support vector machine with radial basis function kernel yields the highest accuracy which is 88% than other classifiers. The support vector machine with radial basis function kernel with db8 could be employed in determining the dyslexic children’s levels objectively during writing.