EEG Channel Selection for AR Model Based ADHD Classification

As of today, diagnosis of ADHD is highly dependent on subjective observations, which has motivated researchers to investigate quantitative methods for the discrimination of ADHD and Non-ADHD subjects using EEG data. The goal of the effort reported here is to classify subjects with high accuracy, as well as to do so based on a select few channels. By making use of AR model features, several classifiers were found to achieve high performance; accuracy above 90% for a K Nearest Neighbor classifier and Area Under the Curve over 0.98 at Equal Error Rate below 0.05 for a Gaussian Mixture Model-Uniform Background Model classifier based on combinations of as few as 2 and 3 EEG channels.