Temporal Attention and Convolutional Tokenization for Interpretable EEG-Based ADHD Identification in Children

Attention Deficit Hyperactivity Disorder (ADHD) is a prevalent neurodevelopmental condition commonly assessed through clinical interviews, behavioral observation, and rating scales. Although electroencephalography (EEG) has emerged as a promising complementary tool for ADHD assessment, robust, subject-independent classification remains challenging due to inter-subject variability, limited datasets, and the need for interpretable computational models. This work introduces EEG-TACT, a compact end-to-end deep learning architecture for identifying ADHD subjects from EEG epochs. The proposed model integrates an EEGNet-inspired convolutional embedding, a Transformer encoder operator, and an attention-based pooling mechanism. Together, these components capture local spatiotemporal EEG patterns, contextual temporal dependencies, and task-relevant latent representations. EEG-TACT was evaluated on a publicly available EEG dataset using strict, subject-independent stratified group partitions, ensuring no data leakage across subjects in the training, validation, and test subsets. Learned temporal filter responses, class-conditioned self-attention maps, and latent-space projections provide model interpretability. An ablation study quantifies the contribution of each architectural component. Performance analysis includes evaluation at the fold, subject, and epoch levels, together with statistical significance comparisons against representative state-of-the-art architectures. EEG-TACT achieved competitive performance among the contrasted models, reaching subject-level accuracy of 87.5%, recall of 96.0%, and precision of 82.8%, while requiring only a few thousand trainable parameters. By exhaustively repeating the initialization, the proposed model demonstrated improved labeling reliability and achieved the best average ranking among the evaluated architectures. The reported results therefore support evidence that EEG-TACT provides a compact, stable, and interpretable model for EEG-based ADHD identification under subject-independent evaluation settings. They also motivate further validation on larger, multi-site, and medication-controlled datasets.