Identifying gambling addiction using logistic regression and naive bayes

Abstract

The problem considered in this study originated from an online gambling company which suggested the task of identifying customers suffering from gambling addiction. Classification algorithms are the obvious tools to use in this problem, and these exist in abundance using diverse statistical and computational techniques. One of the recently developed techniques is the naive Bayes. This has been touted as a well-performing algorithm with superior classification abilities. In this dissertation, we are using it and comparing it with logistic regression, a theoretical problem which has stimulated quite a bit of interest in the literature. Relevant statistical decision and estimation theories are considered. The Bayesian method of estimation has been used for the regression part. Using anonymous records of the company's customers, logistic regression, and Gaussian naive Bayes yield similar results though logistic regression performs slightly better in predictive tasks than Gaussian Naive Bayes.