Enhanced Machine Learning Approach to ADHD Classification Using Actigraphy Data

Attention Deficit Hyperactivity Disorder (ADHD) diagnosis remains challenging due to its heterogeneous presentation and reliance on subjective assessments. This study investigates the potential of actigraphy data combined with advanced machine learning techniques to classify ADHD status. We collected daily activity data from 45 participants (23 with ADHD, 22 without ADHD) from the OBF-Psychiatric dataset and extracted novel features capturing temporal patterns, activity transitions, and circadian rhythm characteristics. Multiple classification models were evaluated, with Support Vector Machines achieving the best performance (F1 score: 0.779). Feature importance analysis revealed that day-night activity transitions, activity burst rates, and established clinical scales were the most predictive indicators. Despite the modest cohort, actigraphy successfully classified ADHD and revealed promising candidate biomarkers, providing valuable insights that establish a foundation for further validation in larger studies.