Background: Attention deficit hyperactivity disorder (ADHD) is an isogenous pattern of hyperactivity, impulsivity, and inattention, resulting in disorders like anxiety, disability in learning, and depression. The electroencephalogram (EEG) signals are a valuable source for early detection of ADHD. However, EEG’s non-linear and non-stationary nature makes its direct analysis very difficult. Method: Different rhythms of EEG offer a robust solution for the automatic detection of ADHD. Therefore, a novel variational mode and Hilbert transform-based EEG rhythm separation (VHERS) is developed. The instantaneous frequency envelops (IFE), and instantaneous amplitude (IA) are extracted using variational mode decomposition and Hilbert transform. The delta, theta, alpha, beta, and gamma rhythms are constructed from the corresponding IFE and IA. Different entropy-based features are evaluated from the rhythms, selected using statistical analysis (mean and STD), and classified using multiple techniques. Results: The proposed VHERS has obtained the highest performance of 100% sensitivity, 99.95% accuracy, a specificity of 99.89%, Cohen’s Kappa of 99.9%, the precision of 99.91%, F-1 score of 0.999, Mathews correlation coefficient of 99.9%, and area under the curve of 99.95%, respectively using a sigmoid kernel of extreme learning machine classifier. Conclusion: The performance shows that the delta rhythm has provided more insight into ADHD and NC EEG signals. The degraded performance for gamma rhythm is due to the overlapping nature of the ADHD and NC EEG features. The proposed VHERS model can help experts to detect ADHD in real-time scenarios.