Abstract

Objectives: The factors that affect students' alcohol use behaviors were examined by logistic regression analysis and artificial neural networks and the efficiency of these methods in identifying alcohol users and non-users was compared using the receiver operating characteristics (ROC) curve method.

Study Design: Graduate students of 1-4 years in Trakya University Medical Faculty (2003-2004) were administered a questionnaire to predict their alcohol use behaviors and were assessed with the Frontal Lobe Personality Scale.

Results: Logistic regression analysis showed that the following variables highly affected alcohol use behaviors of the students: visiting bars, discos or cafes in their spare time (OR=1.920; p<0.05), the importance of religion (OR=0.454; p<0.001), the number of alcohol-user friends (OR=2.441; p<0.001), insistence of friends on taking alcohol (OR=1.557; p<0.01), and impulsiveness (OR=1.826; p<0.001). Comparison between logistic regression analysis and artificial neural networks showed no differences in terms of the areas under the ROC curves of hyperbolic tangent-hyperbolic tangent function and hyperbolic tangent-logistic function artificial neural networks, but these models showed statistically larger areas than the other models.

Conclusion: It may be necessary to take into account the advantages and disadvantages of artificial neural networks and logistic regression in classification and modelling, and to use artificial neural networks to eliminate insignificant variables of logistic regression analysis.