Attention-driven deep learning framework for EEG analysis in ADHD detection.

Abstract

Attention Deficit Hyperactivity Disorder (ADHD) is a neurodevelopmental disorder that affects cognitive functions such as attention, impulse control, and executive functioning. Electroencephalography (EEG) has been widely explored as a noninvasive method for identifying abnormal brain activity patterns associated with ADHD. This study proposes an Attention Module-Based Fused Deep Convolutional Neural Network (AM-FDCNN) to enhance the accuracy of ADHD detection using EEG signals. The model integrates Channel Attention Module (CAM), Spatial Attention Module (SAM), and Position Attention Module (PAM) to selectively focus on critical EEG features, improving classification performance. The dataset, sourced from IEEE DataPort, includes EEG recordings from children diagnosed with ADHD and a control group. The proposed model achieves 97.60% accuracy with an 80-20 training split and 95.12% accuracy with 10-fold cross-validation, outperforming existing machine learning models such as CatBoost, SVM, Random Forest, and Deep CNN. The results indicate that the AM-FDCNN model significantly enhances ADHD detection accuracy, making it a promising tool for clinical and educational applications.