Autism Spectrum Disorder is a neuro-developmental disorder which is usually present when the child is born. But detection of ASD is not possible till the child is around 1-2 years old. Electroencephalography (EEG) can be used to monitor the brain activities of all human beings. As a result, it can be used to detect ASD at an earlier stage. In this study, using Artificial Neural Network EEG signals of children with ASD and non-ASD were classified. Six algorithms of feature extraction namely AR (Autoregressive) Burg, AR Modified Covariance, AR Covariance, AR Yule-Walker, Levinson Durbin Recursion, and Linear Prediction Coefficient algorithm were utilized for feature extraction from pre-processed signals. Six neural network classifiers such as Feed Forward Neural Network, Pattern Recognition Neural Network, Cascade Feed Back Propagation Neural Network, Layered Recurrent Neural Network, Elman Recurrent Neural Network and Nonlinear Autoregressive Network with Exogenous Inputs Neural Network(NARX) were utilized to classify Autistic and non-Autistic subjects. From the study it is proved that NARX neural network with AR Burg features are best to develop the system with a classification accuracy of 96.74% and a maximum bit transfer rate of 6.74 bits/sec.
Volume 11 | 01-Special Issue
Pages: 794-801