A multi-dataset exploratory framework for understanding digital behavior and substance use risk profiles

Area of Science:

• Public Health
• Behavioral Science
• Digital Health

Background:

• Substance use is a complex public health issue influenced by behavioral triggers and digital interactions.
• Understanding the interplay between digital behavior, psychological states, and substance use is crucial for effective interventions.

Purpose of the Study:

• To investigate how demographic factors, psychological states, and digital engagement patterns shape substance use behaviors.
• To develop and evaluate machine learning models for predicting substance use indicators and digital behavioral patterns.

Main Methods:

• Quantitative analysis of NHANES and Kaggle social media psychology datasets.
• Machine learning models (Random Forest, XGBoost, AdaBoost, SVR, Logistic Regression) with hyperparameter tuning.
• Supplementary survey data (N=236) for qualitative interpretation.

Main Results:

• Nonlinear relationships identified between social media engagement, anxiety, and loneliness.
• Anxiety scores plateaued at higher digital engagement levels, indicating qualitative interaction importance over duration.
• Hyperparameter tuning improved machine learning model predictive performance.

Conclusions:

• Digital engagement patterns are vital alongside traditional factors in substance use research.
• Findings support platform-specific digital well-being strategies and nuanced behavioral modeling.
• A multi-source evidence framework can enhance behavioral risk profiling and prevention systems.