Using Classifiers to Identify Binge Drinkers Based on Drinking Motives

Abstract

A representative sample of 2,844 Dutch adult drinkers completed a questionnaire on drinking motives and drinking behavior in January 2011. Results were classified using regressions, decision trees, and support vector machines (SVMs). Using SVMs, the mean absolute error was minimal, whereas performance on identifying binge drinkers was high. Moreover, when comparing the structure of classifiers, there were differences in which drinking motives contribute to the performance of classifiers. Thus, classifiers are worthwhile to be used in research regarding (addictive) behaviors, because they contribute to explaining behavior and they can give different insights from more traditional data analytical approaches.

Keywords :

• classifiers
• nonlinearity
• identifying binge drinkers
• drinking motives

Notes

1 All classifiers shared a confidence interval of ±5% with 95% confidence. The percentage of correctly classified participants over each of 10 repeated tests on subsets of the data had a standard variation of up to 2%.

2 Con1 = 5, Con2 = 4, Con1 = 4, Con4 = 5, Con3 = 5 (conform description in Appendix)