A machine-learning approach for predicting success in smoking cessation intervention

Abstract

Smoking is one of the significant avoidable risk factors for premature death. Most smokers make multiple quit attempts during their lifetime but smoking dependence is not easy and many people eventually failed in smoking quit. Therefore, predicting the likelihood of success in smoking cessation intervention is necessary for public health.In this paper, we analyzed the smoking cessation intervention dataset conducted from the Korea National Health and Nutrition Examination Survey (KNHANES) 2009 to 2017. Accordingly, the chi-square test used to filter relevant and significant features, thus the multivariate analysis was used with logistic regression. In essence, age, education, and frequent alcohol use are important predictors in smoking cessation success. Furthermore, the lowest level of subjective health status has increased the likelihood of unsuccessful smoking cessation.In terms of the class imbalance problem, we have employed an efficient Synthetic Minority Over-sampling Technique (SMOTE) in order to generate new synthetic records. In the current study, we compared the SMOTE regular and borderline-1 techniques with 3, 5 and 7 number of nearest neighbors, respectively. Subsequently, we evaluate the success prediction model of smoking intervention using Naive Bayes (NB), logistic regression (LR), multilayer perceptron neural network (MLPNN), random forest (RF) and gradient boosting trees (GBT) classifiers, as well as classifier performance has evaluated by precision, recall and F-measure.Our result demonstrated that NB with SMOTE borderline-1 (K=5) outperformed the precision of 0.8701. Meanwhile, RF with SMOTE borderline-1 (K=5) performed of 0.8766 and F-score of 0.8476. On the contrary, However, LR presents the lowest F-score as SMOTE regular (K=3) of 0.6726 and borderline (K=3) of 0.6700 in experimental comparison result.In addition, a combination of statistical and machine learning techniques is supposed to be helpful tools in the decisions of smoking cessation intervention and public health domain.