Smartphone movement data can reliably predict smoking lapses and cravings to enable timely smoking cessation support

How your phone can help you quit smoking

Many people now carry a powerful bundle of sensors in their pockets without giving it a second thought: their smartphone. This study asks a simple question with big personal stakes: can the tiny movements your phone already records quietly signal when you are about to feel a strong urge to smoke or slip back into old habits, so that help can reach you in time?

From smoking triggers to silent signals

For years, researchers trying to help smokers quit have focused on obvious triggers such as where someone is, who they are with, or what time of day it is. These cues are useful but imperfect, and they often rely on people reporting their own cravings and slips, which can be inaccurate or incomplete. Smartphones and wearables have changed the game by passively collecting detailed information about how we move, but this rich stream of motion data has rarely been used to understand smoking behaviour outside the lab.

A simple app and months of real life data

The research team asked 17 daily smokers in the United Kingdom to install a basic app on their own phones. For two weeks, participants pressed a single button in the app every time they smoked. Then they attempted to quit, and for three months they logged any lapses and rated their cravings when they were strong, or at least once a day. Throughout both phases, the app silently recorded motion from three built-in sensors: the accelerometer, gyroscope and magnetometer, plus light level and time of day. The phone could be carried however people liked and was not strapped to any particular body part, making the data reflect natural everyday life rather than lab conditions.

Teaching a model to recognise risky moments

To see whether these subtle movement patterns could foretell smoking-related events, the scientists chopped the sensor recordings into non-overlapping five minute slices and labelled each slice as smoking, craving or neither based on the app reports. They then compared several modern artificial intelligence models designed to spot patterns in time series data. The best performer combined two deep learning techniques in a stacked pipeline. When fed only the movement data from the three motion sensors, this model correctly predicted whether a smoking event would occur in the next five minutes about 85 percent of the time during the pre-quit period. It also reached around 77 to 78 percent accuracy in flagging high-craving moments and lapses over the following three months.

Patterns that work across different people

One important question was whether these movement patterns were unique to each person or shared across smokers. To test this, the team trained the model on the pre-quit data from all but one participant, then asked it to predict that remaining person’s lapses and cravings during the quit period. Rotating through the group in this way, the model still performed well: on average it could distinguish high-risk from low-risk five minute windows with a score that indicates good separation between the two. It was particularly strong at predicting cravings, suggesting that subtle changes in how people move before they even touch a cigarette may be a reliable warning sign.

Why tiny movements may matter more than time and place

Perhaps surprisingly, the movement sensors outperformed more traditional predictors such as time of day or light level, which have often been treated as key clues in smoking research. Time of day on its own was the weakest predictor. The magnetometer, which can be influenced by environmental factors, did contribute useful information but only slightly boosted performance when added to the accelerometer and gyroscope. Because the phones were used freely, with no fixed carrying position, it is unlikely that the model simply learned particular places or postures. Instead, it seems to capture broad, repeatable patterns of everyday motion that tend to precede cravings and slips, even when people themselves are not aware of them.

What this could mean for people trying to quit

For someone trying to stop smoking, the most dangerous moments are often brief, when urges spike and it is easy to reach for a cigarette before support arrives. This study shows that motion data already collected by ordinary smartphones can be turned into a kind of early warning system that spots these moments a few minutes in advance, without needing GPS location or detailed personal history. In the future, such models could power quit-smoking apps that quietly watch for risky patterns in your movements and step in with well-timed encouragement, tools or contacts when you need them most, and the same idea may eventually help with other health-related habits as well.

Citation: Abo-Tabik, M., Costen, N. & Benn, Y. Smartphone movement data can reliably predict smoking lapses and cravings to enable timely smoking cessation support. Sci Rep 16, 15719 (2026). https://doi.org/10.1038/s41598-026-49611-y

Keywords: smoking cessation, smartphone sensors, craving prediction, deep learning, digital health