Data mining classification techniques: an application to tobacco consumption in teenagers

This study is aimed at analysing the predictive power of different psychosocial and personality variables on the consumption or non- consumption of nicotine in a teenage population using different classifica- tion techniques from the field of Data Mining. More specifically, we ana- lyse ANNs - Multilayer Perceptron (MLP), Radial Basis Functions (RBF) and Probabilistic Neural Networks (PNNs) - decision trees, the logistic re- gression model and discriminant analysis. To this end, we worked with a sample of 2666 teenagers, 1378 of whom do not consume nicotine while 1288 are nicotine consumers. The models analysed were able to discrimi- nate correctly between both types of subjects within a range of 77.39% to 78.20%, achieving 91.29% sensitivity and 74.32% specificity. With this study, we place at the disposal of specialists in addictive behaviours a set of advanced statistical techniques that are capable of simultaneously pro- cessing a large quantity of variables and subjects, as well as learning com- plex patterns and relationships automatically, in such a way that they are very appropriate for predicting and preventing addictive behaviour.