Classification of the factors for smoking cessation using logistic regression, decision tree & neural networks

Abstract

The idea of saving human lives using machine learning algorithms has pervasive impression. This study has targeted the most significant cause of death and human health deterioration. In this paper we have collected data from different smokers having different smoking status. Several studies have tried to collect the factors but none of them have identified the factors that can lead to quitting this habit. We present a novel approach where machine learning algorithms are used for identification of the factors for smoking cessation. Due to their high classification rate, the used algorithms are Logistic Regression (LR), Decision Tree (J48) and Artificial Neural Network (ANN). Data set is generated by combining different type of sources, in order to identify factors that can classify the status of the smoker as smoker, former smoker and non-smoker. 33 Attributes were used with 10000 instances, class problem was divided in three classes Smoker, Non-Smoker and Former Smoker. The significance of an attribute obtained using the odds ratio, and factors were identified using classification with 10 fold cross validation combined with 1-1 and 1-against all method. The classification rate was highest for Logistic Regression-95.77%, for predicting the class value FSMKR, odds ratio for attribute L-STATUS and B-P-&-HYP have the highest value in predicting the class. Smokers who are living with their parents and not having blood pressure and Hypertension tend to show the cessation behavior. Attribute SMK-MORNING exhibits the pattern of smokers desire to smoke in morning which has the highest value for all three algorithms, while ANN shows only correctly classified instances. When ANN is combined with evaluator and search method results were obtained at 95.10% correctly classified instances.