Classifying Daily Life Smoking Events: An Innovative Wearable Hand-Band Solution

The study addresses the intricate challenge of accurately detecting smoking activity amidst various daily activities monitored by a wearable device. Utilizing data from a wrist-wearable accelerometer, the research employs Naïve Bayes (NB), Discriminant Analysis (DA), Decision Trees (DT), Support Vector Machine (SVM), K-Nearest Neighbour (KNN), and Ensemble Classifiers as machine learning models. A cost-effective wrist-wearable device is designed to collect raw data from (n=28) volunteers engaging in activities for approximately total average of 160±20 minutes. In order to extract time-domain and frequency-domain features from streaming raw sensor data, a sliding window and segmentation method is used. Classification models for multi-class activities (smoking, walking, upstairs, downstairs) were developed and validated employing a leave-one-subject-out (LOSO) approach. The achieved predictive accuracy reached up to 92.0%, with an F1 score of 93.9%. This research paves the way for real-time, accurate smoking detection through wearable devices, offering healthcare professionals timely intervention opportunities for smoking cessation. The framework's applicability extends to diverse preventive healthcare scenarios and the real-time detection of various activities of interest.