Application of Clustering, Logistic Regression and Decision Tree Induction on EGM Data for Detection and Prediction of At-Risk and Problem Gamblers

Abstract

The use of data mining techniques for problem gambling behaviour analysis has huge potential to offer players protection and to reduce the risk of gambling-related harms. In this thesis, we apply three data mining models—clustering, logistic regression and decision tree on one month EGM player data to separate players into different groups, identify which gambling behaviour are highly associated with gambling addiction, and derive predictive rules for predicting potential at-risk and problem gamblers. We consequently separated all players into four groups—non-problem gambler, low-risk gambler, moderate-risk gambler, and problem gambler groups, based on their similar behavioural characteristics. Three behavioural indicators and four best predictive rules are finally obtained to predict at-risk and problem gamblers. It is hoped that this thesis will provide a useful resource for EGM manufacturers to redesign their machines to avoid risky and problem gambling behaviour.