Modelling of Smoking Status Using Bio-Signals: A Comparative Study of Machine Learning and Deep Learning Approaches

The tobacco epidemic, which kills over 8 million people a year—more than 7 million of whom die directly from tobacco use and an additional 1.3 million from secondhand smoke exposure—poses a threat to global health. Despite widespread awareness campaigns, tobacco usage remains prevalent, affecting 22.3% of the world's population, with disproportionate impacts on men and women. In conjunction with this, the study explores the significance of spoken communication in human life and its potential for biometric identification, particularly in discerning smokers from nonsmokers based on voice signals. The research aims to develop predictive models to analyze human bio-signals regarding smoking habits. Furthermore, the study delves into related works, highlighting the importance of predictive modelling in addressing smoking-related health risks and evaluating existing research findings in this domain. The paper also presents the research methodology and outlines the experimental results comparing machine learning (ML) and deep learning (DL) approaches in predicting smoking status. Notably, DL models, particularly the DL with Bi-LSTM model, outperformed ML models, achieving an overall accuracy of 77%. These findings highlight the capability of DL methods in accurately predicting smoking status. They also provide vital information for health stakeholders to create effective interventions in addressing the tobacco epidemic.