JMIR Publications

ABSTRACT

Background:

Digital Just-In-Time Adaptive Interventions (JITAIs) can reduce binge drinking events (BDEs: consuming 4+/5+ drinks per occasion for women/men) in young adults, but need to be optimized for timing and content. Delivering just-in-time support messages in the hours prior to BDEs could improve intervention impact.

Objective:

We determined the feasibility of developing a machine learning model to accurately predict future, that is, same-day, BDEs using smartphone sensor data. We aimed to identify the most informative phone sensor features associated with BDEs on weekend and weekdays, respectively, to determine the key features that explain prediction model performance.

Methods:

We collected phone sensor data from 75 young adults (ages 21-25; mean =22.4, SD=1.9) with risky drinking behavior who reported drinking behavior over 14 weeks. We developed machine learning models testing different algorithms (e.g., XGBoost, decision tree) to predict same-day BDEs (versus low-risk drinking events and non-drinking periods) using smartphone sensor data (e.g., accelerometer, GPS). We tested various "prediction distance" time windows (more proximal: 1-hour; to distant: 6-hour) from drinking onset. We also tested various analysis time windows (i.e., amount of data to be analyzed), ranging from 1 to 12 hours prior to drinking onset, because this determines the amount of data that needs to be stored on the phone to compute the model. Explainable AI (XAI) was used to explore interactions between the most informative phone sensor features contributing to BDEs.

Results:

The XGBoost model performed best in predicting same-day BDE, with 95.0% accuracy on weekends and 94.3% accuracy on weekdays (F1 score = 0.95 and 0.94, respectively). This XGBoost model needed 12- and 9-hours of phone sensor data at 3- and 6- hours prediction distance from the onset of drinking, on weekends and weekdays, respectively, prior to predicting same-day BDEs. The most informative phone sensor features for BDE prediction were time (e.g., time of day) and GPS-derived, such as radius of gyration (an indicator of travel). Interactions among key features (e.g., time of day, GPS-derived features) contributed to prediction of same-day BDE.

Conclusions:

We demonstrated the feasibility and potential use of smartphone sensor data and machine learning to accurately predict imminent (same-day) BDEs in young adults. The prediction model provides “windows of opportunity” and identified “key contributing features” to trigger JITAI prior to the onset of BDEs, with the potential to reduce the likelihood of BDEs in young adults.