An Improved Data Representation for Smoking Detection with Wearable Respiration Sensors

Abstract

Within the field of mobile health, cigarette smoking detection using wearable sensors is a key problem with the potential to improve health outcomes by enabling continuous monitoring as well as personalized, adaptive smoking cessation interventions. Prior work on cigarette smoking detection has demonstrated encouraging results using respiratory inductance plethysmography (RIP) sensors and features derived from inferred respiration cycles. In this work, we argue that the step of pre-segmenting respiration data into respiration cycles is prone to error due to noise in the RIP data. We propose an alternative framing of the problem based on peaks in the RIP data and show improvements in smoking detection accuracy.