Prediction of anxiety disorders using a feature ensemble based bayesian neural network

Area of Science:

• Child and Adolescent Psychiatry
• Machine Learning in Healthcare
• Mental Health Diagnostics

Background:

• Anxiety disorders significantly impact youth development and well-being.
• Early identification of anxiety disorders is crucial for effective intervention.
• Existing screening tools like Youth Online Diagnostic Assessment (YODA) collect valuable parent-reported data, but accuracy can be limited by noisy, self-reported symptoms.

Purpose of the Study:

• To develop and evaluate a novel machine learning approach for improving the accuracy of anxiety disorder prediction in youth.
• To address the challenge of noisy, self-reported data in online screening tools.
• To enhance the diagnostic capabilities of the Youth Online Diagnostic Assessment (YODA) tool.

Main Methods:

• Development and evaluation of a feature ensemble based Bayesian Neural Network (FE-BNN).
• Utilized novel datasets of self-reported anxiety disorder symptoms collected via the YODA tool.
• Compared FE-BNN performance against the original YODA diagnostic scoring function and other baseline methods.

Main Results:

• FE-BNN achieved high Area Under the Curve (AUC) scores: 0.8683 for Separation Anxiety Disorder, 0.8769 for Generalized Anxiety Disorder, and 0.9091 for Social Anxiety Disorder.
• The proposed FE-BNN method demonstrated superior performance compared to existing YODA scoring and baseline approaches.
• Results indicate FE-BNN's potential to more accurately prioritize youth for diagnostic interviews.

Conclusions:

• Feature ensemble based Bayesian Neural Networks offer a promising advancement for accurate anxiety disorder prediction in youth.
• The FE-BNN method effectively handles noisy, self-reported data, improving upon existing screening tools.
• Further research into interpretable methods that maintain high predictive accuracy is warranted.