Attention-Deficit/Hyperactivity Disorder (ADHD) is a prevalent neurodevelopment disorder characterized by persistent inattention, impulsivity, and hyperactivity. Conventional procedures for diagnosing ADHD rely heavily on clinical interviews and self-report questionnaires, which are subjective and often lead to inaccurate diagnoses. Recent advancements in neuroimaging and neurophysiological methods, particularly Electroencephalography (EEG), offer potential as objective biomarkers for ADHD diagnosis. However, the complex nature of EEG signals and the heterogeneity of ADHD symptoms present significant challenges in developing reliable diagnostic tools. In this study, deep learning techniques are explored, specifically Long Short-Term Memory (LSTM) networks, to classify EEG signatures for the detection of ADHD in patients. EEG signatures were documented during gameplay using the EMOTIV EPOC+ device, and features related to different EEG frequency bands, particularly the \(\theta /\beta \) and \(\alpha /\beta \) ratios, were extracted for analysis. This study compares the performance of several methods, both with and without segmentation of EEG data. Among the models, Stacked LSTM achieved the decent outcomes, i.e., 99.40% accuracy. These findings suggest that deep learning models are highly effective in analyzing EEG signals for ADHD diagnosis.