Machine Learning Approaches to Identify and Classify ADHD: A Narrative Review with Tabular Performance Synthesis and Human–AI Mapping

Attention-Deficit/Hyperactivity Disorder (ADHD) presents diagnostic challenges due to heterogeneity, comorbidity rates, and reliance on subjective, phenomenological criteria, resulting in misdiagnosis or treatment delays. This structured narrative review with quantitative tabular synthesis, conceptual mapping, and clinical workflow integration employed a sunflower life-cycle metaphor to bridge clinical expertise and machine learning (ML) technologies, while surveying recent empirical studies (2017–2023) to capture methodological variation in ADHD assessment workflows. Ten studies were selected based on relevance to ML applications for ADHD identification and classification, with deliberate representation of diversity in study design, sample characteristics, data modalities, and ML model-type. The method comprised (a) broad interpretive literature searches, (b) extraction of study-level data, and (c) mapping of ML approaches onto a standardized evidence-based ADHD assessment workflow. Analyses included qualitative synthesis of sample characteristics (youth-focused, N = 38–238,696), data modalities (behavioral surveys, EHR, neuroimaging, genetics), ML models (RF, SVM, DNN), performance metrics, phenotype- and genotype-based distinctions; quantitative aggregation of reported performance metrics (accuracy 66–93%, AUC 0.66–0.94); cross-validation practice, and model-level considerations; and tabular summarization of limitations and multidimensional predictors. Syntheses produced comparative tables, a human–AI diagnostic workflow diagram, and explicit alignment of ML applications with each clinical stage to highlight integration points and gaps.