Predicting substance use behaviors with machine learning using small sets of judgment and contextual variables.

Area of Science:

• Psychology
• Neuroscience
• Data Science

Background:

• Substance use disorder (SUD) is defined by impaired control, dependence, social issues, and risky use.
• Previous research has not directly predicted these core SUD behaviors.
• A need exists for objective measures to assess SUD and its severity.

Purpose of the Study:

• To predict SUD-defining behaviors and SUD severity using judgment and contextual variables.
• To identify specific judgment variable profiles associated with SUD.
• To develop a scalable system for addiction assessment.

Main Methods:

• Utilized data from 3476 adults.
• Employed 15 judgment variables from a picture rating task.
• Applied a balanced random forest approach for prediction modeling.

Main Results:

• Achieved up to 83% accuracy and 0.74 AUC ROC for predicting SUD behaviors.
• Demonstrated moderate-to-high prediction accuracy for substance use recency.
• Reached 84% accuracy in predicting SUD severity.
• Identified that higher SUD severity correlates with increased risk-seeking, reduced loss resilience, higher approach behavior, and lower preference variance.

Conclusions:

• Distinct constellations of 15 judgment variables can predict SUD behaviors and severity.
• This approach offers a scalable system for addiction assessment.
• Findings support further research across various addiction types.