EEG feature extraction using wavelet transform for classifying autism spectrum disorder

Abstract Autism spectrum disorder is a neurodevelopmental disorder in which subjects have impairments in communication, social interaction and is diagnosed on the basis of behavioral assessment, which leads to loss of time for early intervention. Electroencephalography (EEG) is an easy to use and non-invasive technique which measures the electrical activity of the brain. In this study, 16 channel EEG signals measured using scalp electrodes were used to classify ASD affected subjects and healthy controls. Time frequency analysis was performed by applying wavelet transform on the signals for feature extraction. A feed forward neural network was trained using the features and raw EEG signals as input and then used for the classification of the signals. The classification accuracies achieved using statistical features and raw EEG signal as input were 52.7% and 80.9% respectively indicating that the better classification accuracy was achieved using raw data for training the network.