Adam Snoll moved to Huddersfield in 2023 to pursue a research degree after growing up in the United States. There he earned his bachelor’s degree in psychology from the Georgia Institute of Technology in Atlanta. Adam’s research specialty is big data analysis, leveraging technology to identify patterns in large, representative samples. After completing his PhD in 2026, Adam plans to pursue an academic career, contributing to research that aims to improve quality of life for future generations.

Abstract

Using machine learning to predict adolescent alcohol use initiation

Substance use initiation before age 12 is associated with increased risk of later substance use and mental health disorders. This study is the first of its kind to apply a machine learning approach to identify predictors of early alcohol initiation across a broad array of predictive domains. Using secondary data from the Adolescent Brain & Cognitive Development (ABCD) study, a cohort of 7,890 alcohol-naive adolescents (ages 9–10) were followed for three years. An elastic net regression model incorporating 117 variables across eight domains – including demographics, peer and parental behaviour, mental and physical health, culture, hormones, and neurocognition – was used to identify significant predictors. Demographics emerged as the strongest predictive domain: lower initiation rates were observed among Black or Mormon participants, or with lower parental education. Higher rates were associated with Jewish and Orthodox Christian affiliations and older age. Other risk factors included sensation seeking, prenatal drug exposure, and school detention. Higher levels of anxiety/depression, school involvement, and intrinsic motivation were associated with reduced likelihood of initiation. These findings underscore the role of socio-demographic and behavioural factors in early alcohol use, with implications for targeted prevention strategies. Results suggest that low-intensity measures are sufficient in identifying youth at risk of substance use initiation. However, future research should examine whether predictors of initiation differ from those of escalated or problematic use.