An Innovative Approach for Extraction of Smoking Addiction Levels Using Physiological Parameters Based on Machine Learning: Proof of Concept

TL;DR

This study introduces a new method to objectively assess smoking addiction levels using physiological signals and machine learning, offering a more reliable alternative to traditional subjective tests.

Contribution

The study proposes a non-invasive, objective approach to classify smoking addiction levels using machine learning and physiological data.

Findings

Machine learning achieved high accuracy (98.75%) in classifying addiction levels from physiological signals.

There is a direct relationship between physiological parameters and smoking addiction levels.

SMOTE and PCA techniques improved model performance and reliability.

Abstract

Objectives: Determining individuals’ addiction levels plays a crucial role in facilitating more effective smoking cessation. For this purpose, the Fagerstrom Test for Nicotine Dependence (FTND) is used all over the World as a traditional testing method. It can be subjective and may influence the evaluation results. This study’s key innovation is the use of physiological signals to provide an objective classification of addiction levels, addressing the limitations of the inherently subjective Fagerström Test for Nicotine Dependence (FTND). Methods: Physiological parameters were recorded from 123 voluntary participants (both male and female) aged between 18 and 60 for 120 s using the Masimo Rad-G pulse oximeter and the Hartman–Veroval blood pressure monitor. All participants were categorized into four addiction groups: healthy, lightly addicted, moderately addicted, or heavily addicted with the help of FTND. The recorded data were classified using Decision Tree, KNN, and SVM methods. SMOTE and class-weighting techniques were used to eliminate class imbalance. Also, the PCA technique was applied for dimensionality reduction, and the k-fold cross-validation method was employed to enhance the reliability of the machine learning algorithms. Results: Machine learning methods, when evaluated using the SMOTE with a (7380×7) sample of physiological signals recorded every 2 s from 123 participants, showed a high recall of 98.74%, specificity of 99.58%, precision of 98.79%, F-score of 98.74%, and accuracy of 98.75%. Also, it is extracted that there is a direct relationship between physiological parameters and smoking addiction levels. Conclusions: The study’s core novelty lies in leveraging non-invasive physiological signals to objectively classify addiction levels, addressing the subjectivity of the Fagerström Test for Nicotine Dependence (FTND). This study provides a proof-of-concept for the feasibility of using machine learning and physiological signals to assess addiction levels. The results indicate that the approach is promising.

Figures