Prediction of attention deficit hyperactivity disorder based on explainable artificial intelligence.

Area of Science:

• Neuroscience
• Artificial Intelligence
• Psychometrics

Background:

• Accurate Attention Deficit Hyperactivity Disorder (ADHD) diagnosis is vital for effective treatment.
• Traditional methods like the WISC-IV provide diagnostic evidence.
• Machine learning (ML) and eXplainable Artificial Intelligence (XAI) offer advanced predictive and explanatory capabilities.

Purpose of the Study:

• To predict ADHD diagnosis likelihood using ML algorithms.
• To provide interpretable insights into ML model decision-making processes.
• To evaluate the utility of ML and XAI in supporting ADHD diagnosis.

Main Methods:

• Utilized a dataset of 694 cases with WISC-IV scores, age, and gender from Spain.
• Employed stratified 10-fold cross-validation to evaluate diverse ML algorithms.
• Applied feature selection (Boruta) and XAI (Shapley values) for model interpretation.

Main Results:

• Random Forest model achieved high performance (ACC=0.90, AUC=0.94, Sensitivity=0.91, Specificity=0.92).
• A reduced set of 8 key WISC-IV variables yielded comparable results to the full feature set.
• Key predictors included GAI-CPI, WMI, CPI, PSI, VCI, WMI-PSI, PRI, and LN.

Conclusions:

• ML models, particularly Random Forest, demonstrate high accuracy in predicting ADHD diagnoses.
• XAI techniques enhance transparency, aiding professionals in understanding diagnostic factors.
• This ML-driven tool supports clinical decision-making in ADHD assessment, not replacing professional judgment.