Researchers turn smartwatches into drunk driving detectors with 88% accuracy

A multinational team of researchers led by Robin Deuber from ETH Zurich has published the first study demonstrating that consumer smartwatches can detect alcohol-impaired driving. In a randomized, controlled test-track experiment with 54 participants, they collected wrist accelerometry and heart rate variability (HRV) data while subjects drove both sober and after consuming alcohol. The team trained two models: a logistic regression with window-aggregated features and a two-tower 1D convolutional neural network (CNN). The CNN outperformed the logistic model, achieving a participant-averaged AUROC of 0.88 for detecting any alcohol intoxication and 0.86 for detecting driving above the WHO-recommended blood alcohol concentration limit of 0.05 g/dL.

The study is notable for three key innovations: (1) it is the first to use consumer smartwatches rather than specialized hardware for this purpose; (2) the data was collected in a real vehicle on a closed test track, not in a simulator; and (3) the models were rigorously tested on unseen participants to assess generalization. The accelerometer data captures subtle changes in movement coordination, while HRV reflects physiological responses to alcohol. Together, these signals allow the CNN to distinguish impaired driving patterns. The researchers suggest that such a system could be deployed via a smartwatch app to provide real-time alerts, potentially reducing alcohol-related traffic injuries and deaths without requiring additional in-vehicle hardware.