AN ARTIFICIAL INTELLIGENCE TECHNIQUE FOR ATTENTION DEFICIT HYPERACTIVITY DISORDER (ADHD) CLASSIFICATION BASED ON ELECTROPHYSIOLOGY (EEG) ON TWO BENCHMARK DATASETS
Lamiaa A. Amar1, Marwa Y. Mohamed5, Sara Gaballa2,Mariam Hossam2, Ahmed.M.Otifi3,Noha Khattab4
1Department of Networks and Distributed Systems, Informatics Research Institute, City of Scientific Research and Technological Applications, Srta-City, Alexandria, 21934, Egypt.
2Faculty of Engineering, Alexandria University, Alexandria, 21544, Egypt
3Department of data science, Faculty of Computers and Data Science, Alexandria University, Alexandria, 21544, Egypt
4School of Science and Technology, University of Canberra, Australia.
5Department of Multimedia and Computer Graphics, Informatics Research Institute, City of Scientific Research and Technological Applications, Srta-City, Alexandria, 21934, Egypt.
Abstract: The diagnosis of attention deficit hyperactivity disorder (ADHD) is characterized by persistent patterns of hyperactivity, inattention, and impulsivity, which significantly impact academic performance, social interaction, and occupational functioning. Treatment is often inconsistent and delayed due to traditional diagnostic methods that rely on subjective clinical assessments. Electroencephalography (EEG) is a non-invasive neurophysiology technique that measures the brain’s electrical activity, providing a deeper understanding of the neurobiology of ADHD. Machine learning to analyze EEG data, identifying subtle, complex electrophysiological patterns in high-dimensional EEG data that are imperceptible to human observation. This combination leads to more accurate and automated ADHD classification, potentially enabling an earlier diagnosis and personalized intervention. This research aims to develop a more generic screening tool objective for ADHD using EEG signals. Two benchmark datasets were utilized with many features and extracted 24 features across multiple domains, including time, frequency, morphological, and non-linear characteristics. The LASSO-LR model was applied to refine feature selection, identifying the most relevant features for machine learning and deep learning algorithms. Furthermore, the fusion of unhealthy groups within dataset2 results in a strong and high classification accuracy. Among several models tested, the K-Nearest Neighbours (KNN) algorithm achieved the highest accuracy in distinguishing ADHD patients from healthy individuals across both datasets. This study contributes to the growing body of research by offering a novel approach, demonstrating the feasibility of machine learning-based methods as potential diagnostic tools for ADHD diagnosis. Due to their efficiency and adaptability, these models may prove especially advantageous for small-scale edge computing applications in the foreseeable future.
Keywords: ADHD, classification, machine learning, deep learning, feature extraction.
VOLUME 10 ISSUE 01 2026: 65 – 78