Abstract
AIM: Attention-Deficit-Hyperactivity-Disorder (ADHD) is a neurodevelopmental-condition characterized by two symptom-domains, inattention and hyperactivity/impulsivity, as per DSM-5. Prior research, indicates conceptual-overlap among symptoms within each domain, potentially compromising the diagnostic utility of symptom structure itself. This structural redundancy has direct implications for evaluation of ADHD-screening-tools, which already show substantial heterogeneity in item-content and focus. While full psychometric-validation is resource-intensive, assessing tool alignment with DSM-5 offers a more practical and clinically relevant alternative. METHOD: Considering these challenges, this study first employed a three-layer-similarity-framework with entropy-based-weighted-combined-score, to investigate intra-domain symptom redundancy. Subsequently, a multi-stage-classification-pipeline, comprising a filtering-layer and machine-learning-classifiers (Random-Forest, Support-Vector-Machine and Logistic-Regression), was trained on DSM-5 ADHD and Non-ADHD (Conduct-Disorder, Major-Depressive-Disorder, Oppositional-Defiant-Disorder) statements, tested on Vanderbilt-preschool-assessment-questionnaire and validated on ADHD-Rating-Scale, Swanson-Nolan-and-Pelham-Rating-Scale (SNAP-IV) and Modified-Checklist-for-Autism-in-Toddlers (M-CHAT), to assess screening-tool's alignment with DSM-5. RESULTS: The results revealed moderate-overlap between symptom-pairs (2 and 5) and (5 and 7) within the inattention-domain, with similarity-scores of 0.62 and 0.58 respectively. The filtering-layer demonstrated high accuracy of 97%, perfect precision and specificity in isolating ADHD symptoms. Among classifiers, Random-Forest achieved the best performance with 92% accuracy, 83% precision, 100% recall and 91% F1-score. Validation with ADHD-Rating-Scale ensured near-perfect classification due to its focused symptom set, while SNAP-IV's inclusion of non-ADHD-items slightly reduced subtype specificity. M-CHAT validation further confirmed the designed pipeline's ability to exclude non-ADHD symptoms, supporting its classification precision. CONCLUSION: The proposed pipeline can be adopted for analyzing strength and limitations of screening-tools, which serve as a catalyst for refinements, ensuring reliability and effectiveness in practical applications.