Prediction of attention deficit hyperactivity disorder using the comprehensive attention test: a large-scale machine learning approach.

Background: The diagnosis of attention deficit hyperactivity disorder (ADHD) relies on comprehensive approaches, including clinical interviews, scales, and neuropsychological assessments. However, the process is often limited by issues of reliability and availability. Objective: This study aims to develop a robust machine learning (ML) model using large-scale data from the Comprehensive Attention Test (CAT) to predict ADHD diagnoses. Methods: A total of 11,429 participants were recruited across South Korea and underwent the CAT. Of these, 7,737 were diagnosed with ADHD, including 6,772 with comorbid psychiatric conditions. Additionally, 850 individuals were included as a normal comparison group. Eight ML models were trained on a balanced dataset and validated using 5-fold cross-validation. Results: The CAT, when combined with the ML model, achieved an accuracy exceeding 0.98 in distinguishing pure ADHD cases from normal comparison groups. Classification accuracy was particularly high when distinguishing ADHD with comorbid externalizing disorders from normal control groups, especially in cases with more severe ADHD symptoms. Conclusion: The findings of this study suggest that the CAT, integrated with machine learning models, could serve as a promising tool for diagnosing ADHD.