1-norm support vector machine for college drinking risk factor identification

Abstract

College student alcohol misuse is a major public health concern. According to a national survey, about 44% of students engage in high-risk drinking activities. This paper presents a machine learning approach to a secondary analysis of data collected in a college drinking study at the University of Connecticut Alcohol Research Center sponsored by the National Institute on Alcohol Abuse and Alcoholism. Existing alcohol studies are deductive where data are collected to investigate a psychological/behavioral hypothesis and statistical analysis is applied to the data to confirm the hypothesis. However, the collected data often carries information beyond the original hypothesis. Our approach aims to discover knowledge from multivariate data collected at a major university campus, which may or may not confirm the original hypothesis and lead to potentially new insights. The proposed machine learning approach can effectively identify risk and/or protective factors for high-risk drinking that can be used to help detect and address the early developmental signs of alcohol abuse and dependence within college-aged students. We demonstrate the use of a statistical feature selection method, 1-norm support vector machine (SVM), to help classify college students as either heavy or low-risk drinkers and simultaneously select the risk factors for heavy drinkers. Results of our experiments are evaluated by several psychologists to delineate risk or protective factors and the interaction among these factors for college drinking behaviors.