Using Machine-Learning Algorithms to Predict Self-Reported Problem Gambling Among a Sample of Online Gamblers

Abstract Studies suggest that algorithms can effectively be used to predict self-reported problem gambling using player tracking data. The present study analyzed a sample of real-world online gamblers ( N = 1,611) who engaged in lottery playing, casino gambling, bingo playing, and sports betting. The data also comprised each player’s actual gambling activity, as well as age and gender, in the 30 days prior to answering the Problem Gambling Severity Index (PGSI). Players who engaged in at least one lottery game 30 days prior to answering the PGSI were less likely to be problem gamblers compared to players who did not play lottery games. For all other game-categories the relationship was reversed. The results also indicated that specific behavioral tracking features—such as the average number of monetary deposits per session, total amount of money bet per day, session length, and casino gambling involvement—were among the most significant predictors of self-reported problem gambling. When evaluating different machine algorithms, logistic regression and random forest emerged as the most effective in predicting self-reported problem gambling. The present study is among the few which predicts self-reported problem gambling using a sample of online lottery players, casino gamblers, bingo players and sports bettors, and provides further empirical evidence supporting the use of machine learning models to identify self-reported problem gamblers based on player tracking data. These findings can inform responsible gambling strategies by enabling operators to identify and intervene before gambling-related problems escalate.