Determining Dominant EEG Channels for Classification of LD and Non-LD Children using Machine Learning Approach

Abstract

Learning disability (LD) occurs when a person struggles to read, write, name words quickly, and spell. Reading and comprehension are challenging for children with LD. Machine learning, image processing, psychology, and brain anatomy are used to classify LD and non-LD children. The present study proposed a model to classify LD children and to identify dominant brain region in the classification of LD and normal children using SVM classifiers. This study uses electroencephalography (EEG) data to identify the dominant electrodes from 19 channels (electrode location) in occipital, temporal, frontal, and parietal areas. This study included 20 LD and 16 non-LD children with an age group of 8-16 years. Processing of raw EEG signals were performed using discrete wavelet transform to extract information from bands-alpha, beta, delta, and theta bands. The classification model uses support vector machine (SVM) with various kernels- linear, quadratic, cubic, and radial basis function. Results showed the dominance of left temporal electrode locations in identifying the LD children with the maximum accuracy of 94.4% using RBF-SVM classifier.