Abstract: Event-related potential (ERP) can be used for EEG feature extraction and classification for
children with attention deficit hyperactivity disorder (ADHD) and normal children. Firstly, the EEG
signals of two kinds of children were collected by the gambling task paradigm. Secondly, the optimal
electrode was selected based on the Pearson correlation coefficient algorithm, and the optimal
electrode EEG signal was preprocessed. Then, time domain features (mean, variance, peak) and
frequency domain features (Theta band power, Alpha band power) of pre-processed EEG signals were
extracted. Finally, traditional classification methods (Support Vector Machine (SVM), Adaptive
Boosting (AdaBoost), Bootstrap Aggregating (Bagging), Linear Discriminant Analysis (LDA), Back Propagation (BP) and combined classifier classification methods (LDA-SVM, BP-SVM) were used to
complete the classification of two kinds of EEG signals. The results demonstrate that the
classification accuracy of traditional BP classifier was up to 80.52% and that of the combined
classifier was up to 88.88%. The combined classification method can improve the classification
accuracy for ADHD children and provide technical support for ADHD neurofeedback rehabilitation
treatment based on the BCI technology.

Key words: event-related potential, Pearson correlation coefficient, gambling task paradigm, EEG
classification, brain-computer interface