Identifying problematic gamblers using artificial and Bayesian neural networks : a binary and multiclass approach

Abstract

Responsible gaming has gained popularity in recent years due to the harmful nature of compulsive online gambling and the increased awareness on the unfavourable consequences arising from this type of gambling. The focus of this dissertation is to apply machine learning techniques to the responsible gaming problem by training models that can predict the gambling-risk of a player, both as a binary and a multiclass classification problem. Artificial neural networks and Bayesian neural networks are used to train the models using data from past gambling sessions, provided by the gaming company LeoVegas Mobile Gaming Group. These models will then be compared depending on the prediction accuracy of new observations. A two-stage binary approach is proposed by making use of three binary models to obtain a multiclass model. Different performance metrics provide different outcomes to which binary model performed best; however, it is evident that the Bayesian neural network models perform significantly better in reducing the false negative cases, which is further shown in the multiclass models. With reference to the multiclass problem, the two-stage Bayesian neural network model performs best in classifying the fundamental class, the one where the players have a high risk of becoming problematic gamblers.