SmokeSift: Unraveling Smoker and Non-Smoker Individuals Through Machine Learning

The global significance of cigarette smoking in public health highlights the necessity of accurately determining individuals’ smoking status for effective healthcare and treatment planning. Previous research’s lack of a prototype for current models poses a significant challenge addressed in this study. This research develops supervised machine learning models using different algorithms; Logistic Regression, Linear Discriminant Analysis, K-Nearest Neighbor, Decision Tree, Gaussian Naive Bayes, Random Forest, Gradient Boosting, AdaBoost, XG Boost, Extra Tree, CatBoost, Light GBM, and Ridge. These models aim to classify 55,692 individuals as either smokers or non-smokers based on 27 different health-related features. After data cleaning and feature selection (comprising Recursive Feature Elimination, Univariate Selection, and examination of Feature Importance), some redundant and less important features were eliminated, such as Urine Protein, Hearing, Dental Caries, Tartar, and Serum Creatinine, leaving 17 features. Following optimization, Random Forest emerges as the top performer, achieving 88.03% training accuracy and 83.29% testing accuracy. Integrated into a Raspberry Pi with a touchscreen interface, it provides real-time smoking status predictions. This demonstrates machine learning’s potential for precise, automated smoking status identification, benefiting medical assessment. Further research could explore additional health determinants to enhance model accuracy.