Data mining techniques for drug use research

Five classifiers techniques are compared; all of them have much better performance (% of correct classifications) than the simplest classifier (more repeated category: drug use/never drug use) in all the compared drugs (alcohol, tobacco, cannabis, cocaine). Nevertheless, alcohol and tobacco have the lower percentage of correct classifications concerning the drug use motives, whereas these use motives have better classification performance when predicts cannabis and cocaine use. When we analyse the specific motives that better predicts the category classification (drug use/never drug use), the following reasons are highlighted in all of them: “pleasant activity” (most frequent among drug users), and “friends consume” and “addiction” (both of them most frequent among never drug users). These results relate to the social dimension of drug use and agree with the statement that environmental context influences adolescent's involvement in risk behaviours. Implications of these results are discussed.