Detection of cigarette smoke inhalations from respiratory signals using decision tree ensembles

Abstract

In this study we explored the ability of ensembles of decision trees to classify hand-to-mouth gestures in order to detect cigarette smoke inhalations. Three subject independent models were constructed using a variety of ensemble techniques: boosting (AdaBoost), bootstrap aggregating (bagging), and Random Forests. Data was gathered during previous studies by extracting features from the signal waveforms of worn sensors. Each hand gesture was associated with either a smoke inhalation or a hand gesture of another type (e.g. eating). Subject as well as group models were trained. For the group models, model performance was evaluated by computing F-score, precision, and recall statistics using a 20-fold leave-one-out cross-validation testing strategy where one subject was held out for evaluation and models were trained on the remaining 19 subjects. For the individual models, models were trained on a single subject and evaluated using 5-fold cross validation. The average and standard deviation of each statistic across all folds were reported.

Feasibility Assessment of Wearable Respiratory Monitors for Ambulatory Inhalation Topography

“…Many calibration techniques, a necessary step for going from chest motion waveform to lung volume waveform, have been developed, assessed, and reported over the years. In comparison, there have been very few reports of work being done to obtain inhalation topography from lung volume waveform [ 68 , 69 ]. The primary challenge is in identifying each breath in the volume waveform.…”

Feasibility Assessment of Wearable Respiratory Monitors for Ambulatory Inhalation Topography

“…Many calibration techniques have been developed, assessed, and reported over the years. In comparison, there have been very few reports of work being done to obtain inhalation topography from lung volume waveform [69,70]. The key step in obtaining inhalation topography from lung volume waveform is identifying the start and end of an inhale and the start and end of an exhale.…”

Wearable Sensors for Monitoring of Cigarette Smoking in Free-Living: A Systematic Review

“…Signal Processing and Pattern Recognition: Several algorithms were proposed in a series of publications [102,103,104,105] based on the combination of RIP breathing and Proximity sensors of the PACT system. A comparison of these detection approaches is provided in Table 14.…”