An Explainable Intelligence Driven Query Prioritization Using Balanced Decision Tree Approach for Multi-Level...

Area of Science:

• Computational Psychology
• Artificial Intelligence in Healthcare
• Machine Learning for Mental Health

Background:

• Human emotions significantly impact psychological health, with negative emotions exacerbating disorders like anxiety, stress, and depression.
• Existing computational methods for psychological disorder prediction often lack transparency, offering a "black-box" view of uncertainty.

Purpose of the Study:

• To develop a novel, explainable predictive model for multi-class psychological risk recognition.
• To provide an accurate and interpretable interface for understanding psychological disorder prediction.

Main Methods:

• Utilized standard questionnaires as a dataset and introduced Q-Prioritization to refine feature selection.
• Employed a novel balanced decision tree method with repetitive oversampling for model training and testing.
• Integrated permuted feature importance, contrastive explanation, and counterfactual methods for model interpretability.

Main Results:

• Achieved an aggregated accuracy of 98.25% in psychological disorder prediction.
• Reported high performance metrics: mean precision (0.98), recall (0.977), and F-score (0.979).
• Demonstrated the critical role of Q-Prioritization, with accuracy dropping to 90.25% without it; observed a low error rate of 0.026.

Conclusions:

• The proposed multi-level psychological disorder predictive model offers significant improvements in accuracy and explainability.
• The model can serve as a valuable assistive tool for medical experts in mental health treatment.
• Explainable AI techniques enhance the clinical utility of predictive models for psychological disorders.