<p>Various machine learning and deep learning approaches have been proposed for diagnosing attention deficit hyperactivity disorder (ADHD). Most of these studies have employed traditional methods, such as support vector machines, neural networks, extreme learning machines, etc., which still fall short of achieving the desired accuracy rate in ADHD classification, highlighting a need for more effective strategies. Fuzzy logic and metaheuristic-based approaches have emerged as promising strategies for improving classification performance. This study proposes an efficient and robust hybrid method that integrates an adaptive neuro-fuzzy inference system (ANFIS) with puma optimization and incorporates a recursive feature elimination technique for efficient feature selection to enhance learning performance. The proposed approach can (a) utilize fuzzy-based inferencing capabilities with a structure of the neural network, (b) prevent overfitting, and (c) solve the binary and multi-disease classification. This study uses the benchmark personal characteristic data of the ADHD-200 data set to train and evaluate the proposed model. The performance of the ANFIS classifier has been compared with several conventional machine learning classifiers, such as random forest classifier, K-nearest neighbor, and gradient boosting classifier etc. The proposed model achieves an accuracy of 99.40%, a sensitivity of 99.11%, and an F1-score of 98.10%, surpassing previous state-of-the-art ADHD studies. It can provide great assistance to medical experts to correctly classify ADHD patients and healthy individuals. The performance of the proposed model is further validated using four well-known disease data sets. The results demonstrate that the proposed model can adapt to different data sets and be used across various domains.</p>