Abstract
OBJECTIVE: Attention Deficit Hyperactivity Disorder (ADHD) is a neurodevelopmental condition with challenges in timely and accurate diagnosis. This study evaluates the effectiveness of combining electroencephalogram (EEG) data with machine learning techniques to enhance ADHD diagnostic accuracy. METHODS: A total of 168 participants, comprising 107 ADHD and 61 neurotypical (NT) individuals, were assessed using the Kiddie Schedule for Affective Disorders and Schizophrenia Present and Lifetime Version Korean Version (K-SADS-PL-K). EEG data from 19 channels were analyzed across five frequency bands: delta (1-4 hz), theta (4-8 hz), alpha (8-12 hz), beta (12-30 hz), and gamma (30-51 hz). The Extreme Gradient Boosting (XGBoost) classifier was employed for classification, and Leave-One-Subject-Out (LOSO) cross-validation was used to ensure model robustness. RESULTS: Data augmentation through 30-second segmentations generated 2434 EEG segments for ADHD and 1060 for NT. The XGBoost model achieved a test accuracy of 90.81% and an F1-score of 0.9347. Feature importance analysis using SHAP (SHapley Additive exPlanations) values identified middle beta frequency features, particularly from the O1 electrode site, as significant contributors to classification. CONCLUSION: EEG-based machine learning models, such as the XGBoost classifier, show potential as non-invasive tools for ADHD diagnosis, offering high accuracy and interpretability. The novelty of this approach lies in combining SHAP analysis with data augmentation techniques and LOSO cross-validation, ensuring both explainability and robust generalizability. Future research with larger datasets and diverse populations is recommended to validate findings and explore clinical applications.