EEG Based Classification of Learning Disability in Children Using Pretrained Network and Support Vector Machine

Learning disability (LD) is a commonly acknowledged neurological disorder that causes learners to struggle to decode, read, and write and in performing a mathematical task. If a child's learning disabilities go untreated, they may develop lifelong social and emotional issues, which may affect their future success in all spheres of their lives. With the help of early intervention, we can bridge the gap between generally normal developing and children with LD. Electroencephalogram (EEG) can be used to investigate the electrical activity of the brain in order to automatically detect and recognize the disability in young children. This paper proposes a model to classify from the rest EEG signals of normal and LD children. Before feeding the 19-channel EEG signal to convolution neural network (CNN), it was preprocessed, segmented and converted into spectrogram from the alpha, beta, delta and theta bands extracted using wavelet decomposition. The two different modalities proposed in this work were: (1) using pre-trained network for transfer learning approach (2) pretrained network to extract the image features and classification with the help of support vector machine (SVM). In this experiment, the networks like Alexnet, VGG16, and Resnet-18 were compared to compute the results of both the modalities. The highest classification accuracy of 98.3% was obtained using image features extracted from Alexnet and classifying it further using SVM. The use of pretrained network for extraction of image features approach resulted in increased accuracy. The results of the comparison showed that feature-based technique outperformed traditional CNN approach and it may be used for the development of intelligent automated diagnosis system.