This study examined the effectiveness and ease of predicting problem gambling using machine learning. The researchers proposed a novel approach that may be a more objective way to identify people with problem gambling and predict their behaviour. Data on all incoming and outgoing transactions made between 2012 and 2013 on the gambling platform Satoshidice were collected. The data were separated into five 21-day periods. The researchers analyzed each person’s bets over a 10-day period.
Two clusters of people were identified based on their gambling behaviour in the first three days after their first bet: casual and intensive. The intensive group gambled more frequently and placed more bets than the casual group. The intensive group often bet multiple times a day. Both groups often placed “balanced” bets, where they had about 50% chance of winning. But the intensive group had a wider range in terms of placing risky bets and the size of their bets. These key findings were used to create prediction models to identify whether people would or would not place a bet in the 4th to 10th day after the first day they bet. The findings show that machine learning models can be used to identify people with problem gambling in real time.
