ADHD Subtype Diagnosis for Multiclass Imbalanced Augmented EEG Data

Attention-Deficit/Hyperactivity Disorder (ADHD) is a neurodevelopmental condition that impacts an individual’s emotional state, actions, and ability to acquire knowledge. The diagnosis is challenging due to the different characteristics of its distinct subtypes: inattention, hyperactivity, and impulsivity. ADHD has been diagnosed using a range of techniques, such as fMRI and behavioural evaluations. Electroencephalography (EEG) has garnered interest due to its capacity to uncover atypical brain activity patterns. Nevertheless, using EEG data in ADHD research poses difficulties due to the limited annotated datasets and substantial inter-individual variations. To address these difficulties, we have investigated various approaches to improve EEG data: Synthetic Minority Over-sampling Technique (SMOTE) variants, Non-deterministic Conditional Tabular Generative Adversarial Network (ND-CTGAN), and EEG-GAN variants. SMOTE variants have generated synthetic samples to address class imbalances, enhancing classification performance. ND-CTGAN has generated synthetic EEG data of superior quality, preserved inter-channel correlations and spectral properties, and improved the classification model’s accuracy, recall, and precision. Statistical analyses have verified that synthetic data closely reflects the distributions and connections seen in real data. This similarity allows for strong comparisons and improves the ability of models. These findings not only enhance our understanding of ADHD but also inspire hope for future studies in the field, demonstrating the potential of advanced augmentation techniques to significantly enhance classification accuracy in EEG research