Perdition of Smoking in Young Adults Based on Machine Learning Methods:A System Medicine Approach

Tobacco use is one of the main leading cause of preventable death. Numerous studies have shown that intervention to postpone or prevent tobacco use can be an effective strategy to prevent smoking (Talluri, Wilkinson, Spitz, & Shete, ۲۰۱۴). Considering the reduced onset age of smoking, this study focused on predicting the usage status of teenage students for further prevention. In this study, we propose a machine learning framework for automatic classification of students to smoker and non-smoker based on questionnaire data. The main set of variables are including psychological (depression and self-efficacy), family, social, attitudinal and belief factors and school policy toward smoking. The results of specificity and negative predictive value of ۹۳% and ۹۸% respectively, show the high performance of Adaboost classifier in predicting and classifying students as smoker or non-smoker. At the next step, using randomized lasso feature selection, the more effective variables for classification were introduced.

Keywords:

System medicine , Smoking , Classification , Machine learning , Adaboost , Feature Selection