Developing decision support system for determining of smoking cessation therapy with deep learning approach

The World Health Organization (WHO) reported that cigarette addiction is the most common addiction in the world, and every year 6 million people die from diseases related to smoking. Smoking cessation rates in the world and in our country are very low and vary according to the individual. For this, physicians should make treatment plans that take individual differences into account while determining the appropriate treatment method for patients. This study introduces computerized clinical decision support systems to assist the treatment professional and a personalized healthcare system that provides personalized patient recommendations based on patient characteristics. 383 patients who applied to the Smoking Cessation Outpatient Clinic of Trabzon Farabi Hospital Chest Diseases Department between 2015-2020 were included in the study. Using the data obtained from these patients, factors affecting smoking cessation success were analyzed using IBM SPSS23 and R package program. Among the factors obtained from 383 patients using the logistic regression inverse elimination method, 12 factors were identified that turned out to be effective in smoking cessation success. The deep learning method, which has been developing in recent years, is used in analyzing medical data and diagnosing diseases. For this purpose, in this thesis, a deep learning model that can predict people's smoking cessation rate of smokers based on individual differences has been created. Convolutional neural network (CNN) based decision support system, which is a method of deep learning approach, has been developed on the Google Colaboratory platform using a data set consisting of 1000 sample and 12 factors. In this data set, a success rate of %79 was obtained in the CNN network where 80% of the data was divided into training and 20% test data randomly. For the best CNN model developed on 200 test data, %88 specificity, %64 sensitivity, and %79 accuracy values were obtained. CNN's method has been compared with 7 different machine classifiers (RF, DT, NB, SVM, LR, KNN and MLP). The CNN method has proven to be effective on other machine classifiers with an accuracy of %79. Thanks to this expert system, smoking cessation rates were classified in patients who applied to the smoking cessation clinic.