Binge drinking in early adulthood: A machine learning approach

Abstract

Binge drinking among young adults (18-25) has been recognized as a public health concern. Considerable variation among drinking behaviors have been found among this group. Several statistical methods are available to identify theoretically and empirically meaningful correlates of binge drinking. The present study evaluated three methods for identifying correlates of binge drinking, comparing logistic regression to two machine learning methods-classification tress and random forests. While each model identified similar correlates of binge drinking-such as propensity for engaging in risky behaviors, marijuana dependence, cocaine dependence, identifying as non-Hispanic white, and higher education-the AUC analysis showed that the random forest analysis more accurately classified positive cases of binge drinking. Random forests modelling of psychosocial data is a feasible approach for identifying correlates of binge drinking behaviors among young adults. Clinical implications are discussed related to screening for binge drinking in behavioral health organizations.

Keywords: Binge drinking; Machine learning; Random forest; Young adults.