Interpretable machine learning approaches for children's ADHD detection using clinical assessment data: an online web...

Area of Science:

• Neuroscience
• Computational Psychiatry
• Pediatric Psychology

Background:

• Attention-deficit/hyperactivity disorder (ADHD) is a common neurodevelopmental disorder in children, characterized by inattention, hyperactivity, and impulsivity.
• Accurate diagnosis and subtype identification are crucial for effective intervention and management.
• Existing diagnostic methods can be time-consuming and subjective, highlighting the need for objective tools.

Purpose of the Study:

• To develop and validate a machine learning model for identifying ADHD and its subtypes in children.
• To ensure the model is verifiable and interpretable for clinical application.
• To create a user-friendly web application for predicting ADHD probabilities.

Main Methods:

• Utilized the ADHD-200 dataset, including demographic, behavioral, and intelligence assessments.
• Trained and validated seven machine learning models, including Random Forest (RF) and Support Vector Machine (SVM).
• Employed SHapley Additive exPlanations (SHAP) for model interpretability and deployed the best model via a web application.

Main Results:

• The RF model achieved an AUC of 0.99 for ADHD identification, while the SVM model excelled in subtype classification with high accuracy.
• SHAP analysis revealed that the ADHD Index and lower IQ scores were significant predictors, aligning with existing research.
• Model interpretability confirmed the identification of key behavioral scale indicators for ADHD.

Conclusions:

• Machine learning models, particularly RF and SVM, demonstrate high accuracy in diagnosing ADHD and its subtypes in children.
• Interpretable AI can support clinical decision-making by providing objective insights into ADHD diagnosis.
• The developed web application offers a practical tool to improve diagnostic efficiency and accuracy in pediatric ADHD assessment.