Leveraging data augmentation for machine learning models in predicting depression and anxiety using the Revised Child...

Area of Science:

• Psychiatry
• Machine Learning
• Public Health

Background:

• 15 million people in Pakistan are affected by depression and anxiety.
• Limited government mental health facilities and psychiatrists necessitate efficient screening tools.
• Urgent need for effective intervention strategies to address the mental health crisis.

Purpose of the Study:

• To utilize machine learning (ML) with the Revised Children's Anxiety and Depression Scale (RCADS) for efficient screening.
• To maximize the use of existing healthcare resources for mental health assessments.
• To facilitate early detection and intervention for depression and anxiety in children and adolescents.

Main Methods:

• A dataset of 89 cases (from 138 initial) with 47 RCADS items was used.
• Feature selection involved Chi-square, Spearman's correlation, and Random Forest-Recursive Feature Elimination.
• Data augmentation and training of six ML algorithms (Decision Tree, Random Forest, SVM, Logistic Regression, Naive Bayes, KNN) were performed, with top models validated on augmented and external datasets.

Main Results:

• Random Forest algorithm achieved up to 81% macro average accuracy, precision, recall, and F1 score.
• Optimal data augmentation was found to be four times the original dataset size.
• RCADS Item 05 showed weak correlation with depression and anxiety evaluation in the study population.

Conclusions:

• The Random Forest algorithm, using 46 features, is a suitable tool for screening anxiety and depression.
• This ML approach shows potential as a decision support system for mental health professionals.
• The study highlights a promising method to improve traditional screening for pediatric anxiety and depression.