Role of personal characteristics data for classification of attention-deficit hyperactivity disorder

Abstract

Diagnosed in millions of children, ADHD is the leading mental health concern in childhood. Several steps and a lot of personal characteristics (PC) are required from various sources for accurate analysis of ADHD and its subtype (Hyperactive (ADHD-H), Combined (ADHD-C), or Inattentive (ADHD-I)). Moreover, there is no standard automatic diagnostic tool to differentiate ADHD, its subtype, and typical developing (TD) using PC data. The present work focused on the development of a machine learning-based automatic diagnostic tool for the classification of TD, ADHD, and its subtypes using PC data that can be helpful for clinicians. In this work, eight datasets (D1 to D8, four balanced and four unbalanced) are constructed from publicly available dataset and three sets of features were built. Five popular classifiers, namely K-Nearest Neighbor (KNN), Logistic Regression Classifier (LRC), Random Forest (RF), Support Vector Machine (SVM), and Radial Basis Function Support Vector Machine (RBSVM), were trained for the datasets. To comprehensively evaluate performance, the evaluation involved ten iterations of a 10-fold cross-validation approach to calculate average classification accuracy, recall, specificity, and F1 score. Gender, IQMeasure, Full4IQ, and Handedness are observed to be relevant for the classification. Overall, it is observed that RBSVM outperformed other classifiers in most cases.