A Study of the ADHD-200 Repository for Attention Deficit Categorization Using Deep Neural Networks

Attention Deficit Hyperactivity Disorder(ADHD) is a socio-behavioral disorder, affecting nearly 1% of people worldwide. Presently, the only traditional approach for diagnosing ADHD is standardized tests which prolong the procedure and raise the expense of care. The goal of this article was to explore the potential of Neural Networks in classifying ADHD patients and controlled children. For this purpose, the study utilizes Personal characteristic Information (PCI) attributes extracted from a sizable collection. ADHD-200 is a multi-site repository that gathered information from several sources including eight different research and clinical institutes. The methodology and results involve two steps: In order to distinguish ADHD from typically developed(TD), a neural network model (NN) was first constructed in Python using the TensorFlow framework. Second, by employing the same repository, the experimental design utilizes the four distinct hidden layer Neural Network models configurations and demonstrated that the Neural Network with configuration(1024-512-256-2) performed best, averaging 75% accuracy. With regard to the ADHD-200 dataset, its accuracy is much greater than any other that is currently known. This study determined the best PCI characteristic for ADHD classification performance. In the future, additional discriminative features such as neuroimaging may enhance performance and ultimately enable the clinical diagnosis of ADHD.