Identifying compulsive gamblers using Bayesian networks

Abstract

The grave consequences suffered by online problem gamblers has led to a growing interest in responsible gambling measures with the intention of preventing players from reaching such a vulnerable state. The focus of this dissertation is to apply statistical machine learning techniques to predict whether a player is most likely a problem gambler or not, whilst identifying which variables were deemed useful predictors of problem gambling. Bayesian networks are implemented on a data set containing historical data obtained from a local medium-sized Malta Gambling Authority (MGA) gambling operator. The models are tested based on a number of goodness of fit measures such as the predictive accuracy and Area Under the Curve (AUC).