LBP-Moment: A Lightweight and Interpretable EEG Framework for Automated ADHD Detection

Attention-deficit/hyperactivity disorder (ADHD) is a common neurodevelopmental disorder. Here we present LBP-Moment, a lightweight and interpretable framework for automated ADHD classification from electroencephalography (EEG). Each 5-s segment was represented by a 256-dimensional local binary pattern histogram and 40 statistical features (IST40). Iterative Chi-square selection reduced the resulting 296-dimensional space to 132 features. A k-nearest-neighbour classifier (k=1; Manhattan distance) was evaluated by segment-level 10-fold cross-validation. The model achieved 97.65% accuracy, a 97.65% F1-score, a Cohen's kappa of 0.9527 and an area under the receiver operating characteristic curve of 0.9762. Ablation analysis showed complementary contributions from LBP and IST40; feature selection removed 55.4% of the features and increased accuracy by 0.35 percentage points. These findings support a transparent, CPU-based feature-engineering approach, but participant-wise and external validation are required before clinical interpretation.