AI-IoT assisted wearable bio-impedance sensor for classification of smoking habits on Fagerstrom scale

Abstract

Understanding smoking habits and their dependency is crucial, as it provides valuable insights into an individual’s respiratory health and lung capacity. We report the development of an IoT-enabled wearable that can measure the bio-impedance of the thoracic region of subjects, transmit the measurements to a remote server using message query telemetry transport (MQTT) protocol, and classify the smoking habits on Fagerstrom scale employing machine learning (ML). The bio-impedance data was collected from 341 smokers post which the dataset underwent calibration, filtering, and feature extraction. The feature-extracted data was labelled using Fagerstrom scale, the scores being calculated using a standard questionnaire. The extracted features were used to train k-nearest neighbour (kNN), random forests (RF), and support vector machine (SVM) classifiers in Python 3.5, using stratified k-fold cross validation. The SVM achieved highest classification accuracy of 97%, followed by RF (96%) and kNN (93%) on Fagerstrom scale with scores ranging between 0–10 highlighting the wearable’s ability to accurately classify smoking dependency, offering a reliable tool for comprehensive respiratory health monitoring. The feature importance results revealed cell membrane capacitance, heterogeneity of tissue and age were most significant features, establishing their importance in lung damage due to smoking. Higher respiratory rates for heavy smokers as compared to light smokers on Fagerstrom scale further established strong dependence of behavioural aspect of smoking habit with lung damage and lung capacity. A key application of this wearable lies in the health insurance sector, where accurate assessment of smoking habits is critical for determining insurance premiums.