USING MACHINE LEARNING TO PREDICT QUIT ATTEMPTS AMONG ADULT CIGARETTE SMOKERS

Cigarette smoking continues to be a national public health concern, leading to serious illnesses like cancer and resulting in premature death. Many interventions have been developed to help cigarette smokers quit. However, in 2018 only roughly 55% of adult smokers made a quit attempt over the past year. The aim of the present study was to demonstrate the application of machine learning and conduct secondary data analysis to explore predictors of self-reported quit attempts among adult smokers. Participants (N = 231; 18 – 69 years old, M = 49.79, SD = 11.42; 52.2% male) had completed baseline and follow-up assessments for an experiment testing a looming vulnerability induction, answering questions about smoking rate, self-efficacy, outcome expectancies, and contemplation status (Haaga et al., 2020). Feature selection was conducted using significance testing and binary classifiers of random forests and support vector machines were chosen. Several methods, including threshold moving and SMOTE, were applied to address class imbalance in the outcome variable. Models were unable to predict quit attempts well in the testing set, with F1 scores ranging from 0.00 to 0.47. Number of previous quit attempts, throat burning, and contemplation status were consistently identified as important features for accurate model predictions. Future research should apply machine learning methods to a larger sample with a wider range of variables previously found to be related to smoking cessation. Further, machine learning may be useful for exploring differences among smokers who have immediate quitting success, compared to those who repeatedly make quit attempts. Machine learning methods have the potential to elucidate the complexity of smoking behaviors and optimize individualized treatment.