Pooled and person-specific machine learning models for predicting future alcohol consumption, craving, and wanting to drink: A demonstration of parallel utility.

Background and Aims: The specific factors driving alcohol consumption, craving, and wanting to drink, are likely different for different people. The present study sought to apply statistical classification methods to idiographic time series data in order to identify person-specific predictors of future drinking-relevant behavior, affect, and cognitions in a college student sample. Design: Participants were sent 8 mobile phone surveys per day for 15 days. Each survey assessed the number of drinks consumed since the previous survey, as well as positive affect, negative affect, alcohol craving, drinking expectancies, perceived alcohol consumption norms, impulsivity, and social and situational context. Each individual’s data were split into training and testing sets, so that trained models could be validated using person-specific out-of-sample data. Elastic net regularization was used to select a subset of a set of 40 variables to be used to predict either alcohol consumption, craving, or wanting to drink, forward in time. Setting: A west-coast university. Participants: Thirty-three university students who had consumed alcohol in their lifetime. Measurements: Mobile phone surveys. Findings: Averaging across participants, accurate out-of-sample predictions of future drinking were made 76% of the time. For craving, the mean out-of-sample R² value was .27. For wanting to drink, the mean out-of-sample R² value was .27. Conclusion: Using a person-specific constellation of psychosocial and temporal variables, it may be possible to accurately predict drinking behavior, affect, and cognitions before they occur. (PsycInfo Database Record (c) 2021 APA, all rights reserved) Public Health Significance Statement—This study utilized both person-specific and between-subjects approaches to predict future drinking, craving, and wanting to drink. Results indicated that both approaches performed well, but may have different applied utility. (PsycInfo Database Record (c) 2021 APA, all rights reserved)