Identifying Marijuana Use Behaviors Among Youth Experiencing Homelessness Using a Machine Learning-Based Framework:...

Area of Science:

• Computational social science
• Public health informatics
• Machine learning applications

Background:

• Youth experiencing homelessness exhibit disproportionately high rates of substance use (69% dependence).
• Structural and social inequalities exacerbate substance use challenges and hinder access to care for this population.
• Existing disparities necessitate innovative approaches for identifying and supporting at-risk homeless youth.

Purpose of the Study:

• To develop a machine learning framework for predicting substance use (marijuana probability) in homeless youth.
• To leverage social media content (posts, interactions) for risk prediction.
• To enable social workers and care providers to proactively identify and assist high-risk individuals.

Main Methods:

• Recruited 133 homeless youth, collected 1-year social media data, and administered surveys.
• Utilized social sharing of emotions and social support theories to identify predictive features.
• Applied natural language processing (NLP) to extract features from social media data for machine learning models.

Main Results:

• Machine learning models achieved an area under the curve of 0.72 and accuracy of 0.81 using only social media data.
• Models demonstrated fairness by evaluating false-positive rates across sex and age segments, avoiding survey data biases.
• Predictive performance was validated without incorporating potentially biased survey information.

Conclusions:

• Social media interactions among homeless youth and their peers are valuable for predicting substance use.
• The developed framework enables efficient resource allocation to vulnerable youth with minimal overhead.
• The approach can be expanded to predict other health behaviors in this population.