Predicting methamphetamine use of homeless youths attending high school: Comparison of decision rules and logistic regression classification algorithms

Methamphetamine use among homeless youths is an increasing problem. School officials and social work practitioners are presented with a classification problem in determining which youth are or are not using methamphet-amines. The purpose of this study is to adopt a machine-learning approach to address this type of classification problem and to compare 2 models (decision rules and logistic regression) for classifying cases into methamphetamine users and nonusers. The selection of predictors in our models was guided by the risk and resilience framework. Logistic regression and decision rules analyses are used to test the models with a subset of data for 2,146 homeless youth who attend high school that was obtained from the 2007–08 California Healthy Kids Survey data-set. Results of logistic regression suggest methamphetamine use is associated with cigarette and marijuana use, having consumed alcohol, and being truant more than once per week. Results of decision rules analysis suggest a youth’s being classified as a methamphetamine user depends on whether the youth has tried marijuana at least once and whether the youth has been truant more than once per week. Moreover, classification as a methamphetamine user also depends on whether a youth has tried marijuana at least once, has not been truant more than once per week, and has tried cigarettes at least once. The logistic regression and decision rules models produce similar—but not identical—results. Our findings highlight the utility of decision rules models as a complement to logistic regression when classification is the goal of a study. Such models can be used to guide social work practice decisions in making informed predictions about client outcomes.