Predicting ADHD from Cognitive Measures
Research Question
ADHD is commonly assessed using self-report symptom questionnaires, but these measures capture only personal subjective evaluation. This project examined whether cognitive and language-related measures could predict ADHD diagnosis independently and whether adding self-reported ADHD symptoms improved prediction.
Approach
A hierarchical modeling framework was developed to compare prediction across different sets of measures, beginning with cognitive and language-related predictors and then incorporating self-reported ADHD symptoms. Multiple interpretable and predictive approaches were compared, including Logistic Regression, Lasso, LDA, QDA, Random Forest, and XGBoost, with emphasis on model generalization, stability, and interpretability rather than maximizing performance alone.
Multiple predictive approaches were compared, including Logistic Regression, Lasso, LDA, QDA, Random Forest, and XGBoost. Models were evaluated with emphasis on generalization, stability, and interpretability rather than maximizing performance alone.
Key Findings
Cognitive and language-related measures demonstrated predictive value for ADHD diagnosis. Adding self-reported symptom measures did not appear to meaningfully improve model performance within this framework, suggesting the possibility that cognitive measures alone may capture information relevant for predicting ADHD diagnosis.
Why It Matters
Cognitive and language-related measures demonstrated predictive value for ADHD diagnosis. Adding self-reported symptom measures did not appear to meaningfully improve model performance within this framework, suggesting the possibility that cognitive measures alone may capture information relevant for predicting ADHD diagnosis.
Transferable Skills
Research design · Measurement selection · Experimental design · Quantitative analysis · Machine learning ·Explainability · Translating data into insights