Understanding predictors of binge-drinking behaviors among college students: a random forests analysis.

Abstract

Objective: This study used a machine learning approach to examine how marijuana use, social connectedness, psychological distress, and demographic characteristics predicted binge drinking (BD) among 31,120 college students from the 2021-2022 Healthy Minds Study. Methods: A random forests analysis was used to assess the influence of predictors of binge drinking, and partial dependence plots and mixed-effects logistic regression were used to determine the direction of each predictor. Results: Findings indicated that marijuana use, older age, lower religiosity, depression, and diminished sense of belonging were among the strongest predictors of binge drinking. Conclusions: This study found that marijuana use, emotional distress, and diminished belonging were associated with higher predicted risk of BD, whereas loneliness showed the opposite pattern. Younger age and higher religiosity were associated with lower predicted risk of BD. Findings may inform risk stratification and campus-level prevention efforts that integrate mental health and substance use services.