A hybrid CNN-machine learning model for anxiety screening with mandala coloring patterns: Feature extraction and...

Area of Science:

• Computational psychiatry
• Digital art therapy analysis
• Machine learning in mental health assessment

Background:

• Traditional anxiety assessments often miss behavioral cues by focusing on self-reports and somatic symptoms.
• Mandala coloring involves visual elements that may correlate with psychological states.
• Digitized art offers potential for objective, quantifiable analysis.

Purpose of the Study:

• To develop and evaluate a hybrid convolutional neural network (CNN) and machine learning (ML) model for classifying anxiety levels using visual features from mandala images.
• To explore the utility of quantifiable visual features (coloring accuracy, transitions, entropy, symmetry) as potential markers for anxiety.
• To assess the model's performance, particularly under class-imbalanced conditions.

Main Methods:

• A dataset of 1044 digitized mandala images was analyzed.
• A hybrid CNN-ML model was employed to extract and classify visual features.
• Stratified 5-fold cross-validation was used to evaluate the Random Forest (RF) classifier's performance.
• Key visual features analyzed included coloring accuracy, color transitions, color entropy, and symmetry.

Main Results:

• The Random Forest classifier achieved an overall accuracy of 84.56% in classifying anxiety levels.
• Model performance was limited under class-imbalanced conditions, showing low sensitivity (0.216) for the high-anxiety group.
• Quantifiable visual features demonstrated potential as exploratory markers.

Conclusions:

• Quantifiable visual features from mandala images can serve as supplementary markers to support anxiety severity classification.
• The developed computational approach offers preliminary evidence for integrating art-based methods with computational modeling for assessment support.
• This hybrid model is not a standalone diagnostic tool but an exploratory framework.