Multimodal Physiological Assessment of Attentional States and Driving Styles in Simulated Driving

48 drivers, 1 h simulator, five physiological signals + eye tracking, attention probes every 2–3 min; with leave-one-subject-out, physiology alone predicted self-reported driving style at balanced accuracy 0.65, and risky drivers drifted from focus to mind wandering twice as often as safe drivers. Authors stress it is not real-time detection.

Mind wandering during driving is one of the major sources of road safety risk, but its expression across drivers of different styles remains unclear. Few studies have examined attentional state and driving style concurrently in a 1 h multimodal simulated driving protocol. We collected one hour of driving, five physiological signals, and eye tracking from 48 participants in a simulator. The driving style was obtained from a Multidimensional Driving Style Inventory pre-questionnaire, and attentional state was probed every 120 to 180 s during the experiment. We analyzed feature-level effects with Mann–Whitney U tests, modeled attentional transitions between adjacent attentional states, and compared seven sensor combinations under six classifiers with leave-one-subject-out validation. The study yields three contributions. First, driving style corresponded to distinct multimodal physiological and gaze signatures, and physiology-only features predicted the style label at a balanced accuracy of 0.65, supporting style as a multimodal trait beyond self-report. Second, attention dynamics differed by driving style, with risky drivers drifting from focus to mind wandering twice as often as safe drivers. Third, the optimal sensor configuration depended on the prediction target and driver subgroup: physiology and eye tracking were the most informative sensors for attentional state, while driving style classification depended primarily on physiological features. These results should not be interpreted as evidence for real-time detection. Instead, they provide empirical evidence for sensor selection and suggest that wearable driver-state estimation should be integrated with vehicle- and context-aware monitoring systems.

Verbatim from the publisher's DOI record (opens in a new tab) · Publisher licence: https://creativecommons.org/licenses/by/4.0/

No results from this paper on ChatPPG leaderboards yet. Browse the leaderboards.