Smoking lapses and cravings can be predicted with high accuracy using smokers’ smartphones’ movements.

Abstract

Decades of research on smoking-behavior had identified triggers that contribute to failed quitting attempts in order to develop effective interventions to support smokers. Triggers may be environmental (e.g., location), social, or internal (e.g. stress). Here, we show that movement data collected from smokers’ smartphones’ sensors (accelerometer, gyroscope and magnetometer) is a better predictor of smoking-behavior. Feeding the movement data into a Deep Learning (DL) model (1D-CNN-BiLSTM), smoking-behavior was
predicted with 85% accuracy within a 5-minute window. This is compared to 63% accuracy when using traditional triggers such as time of the day. Crucially, movement data can be used to predict high-craving and lapses in the 3-months following quitting smoking with similarly high accuracy, even when predictions are made without any personal data (i.e., when the model is trained using only data from other smokers). Findings can be used to transform smoking-cessation smartphone apps, enabling the provision of just-in-time personalized support to those wishing to quit smoking. Importantly, the findings have implications beyond smoking-cessation applications, by revealing that human movements, largely overlooked to date, can be used for, early detection of, and intervention for, health (and other) behaviors, including those that are not genetic or typically characterized by changes in movement.