Using the Random Forest Algorithm on Customer Gambling Data for Predicting Gambling Freezes in an Online Gambling Platform

• Background. Data on gambling behaviors routinely collected on online gambling platforms can be used to detect individuals at risk of developing or having gambling problems. As only data on gambling activity is available on gambling platforms, it is important to find a proxy measure for gambling problems. Temporarily freezing one or several gambling categories has potential to serve this purpose. Aim. To predict gambling freeze in a sample of active users of an online gambling platform one week before the freeze, based on one week of behavioral data tracked on the platform. Method. N = 105 predictors were created, covering total values, frequencies, variations, and trajectories of monetary and time-related gambling involvement, number and type of games played, point in time when gambling occurred, age, and gender. The random forest algorithm was applied to a sample of N = 2618 gamblers (of which N = 1309 freezers), with the sample divided 70/30 into a training and testing data set. Results. The accuracy of random forest applied to the testing data set was 0.615, with sensitivity of 0.543 and specificity of 0.686. The five most predictive variables were current age, age on registration date, average session length, average sum of winnings per session, and total session length. Discussion. The predictive accuracy of the algorithm in the current study was relatively low, suggesting the need for a more suitable target variable. Also, analyzing data collected during a longer period might be needed to create a tool that could be used to identify at-risk gamblers.

Ämnesord

SAMHÄLLSVETENSKAP  -- Psykologi -- Tillämpad psykologi (hsv//swe)

SOCIAL SCIENCES  -- Psychology -- Applied Psychology (hsv//eng)

Nyckelord

Psychology

psykologi

Publikations- och innehållstyp

vet (ämneskategori)

ovr (ämneskategori)