Attention deficit and hyperactivity disorder classification in quantitative EEG signals using machine learning algorithms

Abstract

Attention deficit and hyperactivity disorder (ADHD) classification method as a quantitative observation has been continually improved to assist medical practitioners. Currently, machine learning algorithms such as k-nearest neighbors (KNN), multilayer perceptron (MLP), and support vector machine (SVM) are widely used. This study proposed a feature extraction method for quantitative electroencephalography (qEEG) data derived from the continuous wavelet transform (CWT) to classify children with ADHD versus healthy subjects. Subsequently, this study compared the performance of the classification pipeline before and after the implementation of principal component analysis (PCA) on the features prior to processing with machine learning algorithms. The results revealed that the overall performance of the classifiers consistently improved after the implementation of PCA. The results highlight the varying impact of PCA on classifier performance, with KNN showing an improvement in testing accuracy from 61.84% to 69.21% following PCA implementation, while the other classifiers showed deterioration in performance. These findings suggest that while PCA may be beneficial for some classifiers, its impact on performance varies depending on the specific characteristics of the dataset and the classifier utilized. Moreover, this study provides insight for future implementation of the classification method for ADHD patients across a more specific clinical range of the spectrum.