Identifying Significant Correlates of Adolescent Online Gambling Using A Random Forest Approach: A Cross-Sectional Study

Online gambling is increasing among adolescents. Machine learning methods can be used to identify and rank classification-relevant predictors among a large set of variables. This study identified and ranked factors associated with online gambling among high school students in Canada using a Random Forest (RF) approach. This study used repeat cross-sectional data from Year 6 (2017–18) to Year 8 (2019–20) of the COMPASS study, including n = 103,008 students in grades 9 to 12 from 151 schools in 4 Canadian provinces. Students reported engaging in online gambling in the past 30-days, biological factors (i.e., socio-demographic characteristics), personality factors (e.g., mental health), social factors, and behavioural factors (e.g., movement behaviours, substance use). A RF algorithm identified and ranked factors associated with online gambling. The prevalence of past 30-day online gambling was 3.5%-3.8% across the three years. Across 25 iterations, the RF identified mental health factors (flourishing score, symptoms of depression, emotional regulation, symptoms of anxiety), physical activity time in the past week, and screen time behaviours (time spent texting/watching TV/surfing the Internet/playing video games per day) as factors consistently associated with online gambling in the past 30-days. The average Mean Decrease in Accuracy of the top 10 factors ranged from 47.7 to 34.6. Mental health factors and screen time behaviours were consistently ranked among the top factors associated with online gambling among Canadian adolescents, suggesting that these may be important factors to include in future studies of adolescent online gambling. Future studies should investigate predictors of online gambling using prospective study designs.