Predicting Adolescent Nicotine Vaping: A Machine Learning Approach to Identifying Complex Risk Factors

• In Event: Innovative Applications of Computational Methods

Abstract

This study investigates risk factors associated with adolescent nicotine vaping through an innovative two-stage analytical approach. We combine supervised machine learning techniques with nested logistic regression to analyze data from the Monitoring the Future study (2017-2023, n=72,712). Our six-model machine learning expert system revealed non-linear relationships and interactions that were subsequently validated through nested logistic regression.

Key findings include: (1) a substantial increase in vaping odds (63%) between 2020-2021; (2) negative correlation between marijuana/alcohol use and nicotine vaping, suggesting potential displacement effects; (3) a positive association between educational aspirations and vaping likelihood; and (4) curvilinear relationships between nicotine vaping with cigarette use, academic performance, and social engagement.

These results challenge conventional narratives about risk factors for nicotine vaping. Methodologically, this study demonstrates the value of integrating machine learning with regression to study complex, non-linear relationships in substance use research. Our findings also carry important policy implications, which is discussed in the end of the paper.

Authors

• Siyang Ni, Pennsylvania State University

• Jeremy Staff, Pennsylvania State University