Machine learning can be useful in identifying people with problem gambling. This study assessed the ability of a random forest (RF) model to identify problem gambling among lottery loyalty program members. Ticket data were provided by a state lottery authority in the United States. The lottery authority also invited program members to complete a survey. The survey included the Problem Gambling Severity Index (PGSI) to assess problem gambling. Lottery program members at higher risk of problem gambling (PGSI score of 5 or higher) tended to be younger, less likely to work full-time, and have lower levels of education. They also had lower income levels and were not as likely to be married. Overall, the RF model worked fairly well. But the model’s sensitivity was poor, as it failed to identify people with problem gambling effectively.
