NeuroFusion-ADHD: Hybrid SVM–CNN Framework for Multimodal EEG-Based ADHD Detection

Attention-Deficit/Hyperactivity Disorder (ADHD) needs reliable, objective screening to aid in clinical assessment. However, EEG-based classification is challenging because of the presence of artifacts and high inter-subject variability, and conventional pipelines based on handcrafted spectral/statistical features with classical classifiers (e.g., SVM) may fail to capture complex spatial--temporal patterns. In this work a multimodal framework combining handcrafted EEG features (time domain statistics, band power, theta/beta ratio) and deep features learned by CNN followed by SVM based features for final decision making is proposed. Experiments conducted on a public 19-channel EEG dataset from the data repository of the The Institute of Electrical and Electronics Engineers (IEEE DataPort) (121 children: 61 affected by Attention Deficit Disorder (ADHD), 60 controls; age: 7-12 years old) show better performance for the multimodal model (Accuracy 0.92, Precision 0.91, Recall 0.92) than SVM-only (0.85 accuracy) and CNN-only (0.90 accuracy), while the Receiver Operating Characteristic (ROC)- Area under the Curve These results showed that the combination of interpretable handcrafted descriptors with CNN representations provides a more robust EEG-based detection method of AD/HD than using either method separately.