Key result
Machine learning models predict single-user stress with ~98% accuracy but falter across different users.
Observational (n=4)
Machine learning models can accurately predict upcoming stress levels for individual drivers based on current stress, driving behavior, and road shape, but generalize poorly across different drivers.
Answers are grounded in this paper and cite their sources.
Heart rate sensors may enable real-time driver stress detection; leaves open whether vehicle adaptations reduce cardiovascular risk or crashes.
The levels of stress while driving affect the way we drive and have an impact on the likelihood of having an accident. Different types of sensors, such as heart rate or skin conductivity sensors, have been previously used to measure stress related features. Estimated stress levels could be used to adapt the driver's environment to minimize distractions in high cognitive demanding situations and to promote stress-friendly driving behaviors. The way we drive has an impact on how stressors affect the perceived cognitive demands by drivers, and at the same time, the perceived stress has an impact on the actions taken by the driver. In this paper, we evaluate how effectively upcoming stress levels can be predicted considering current stress levels, current driving behavior, and the shape of the road. We use features, such as the positive kinetic energy and severity of curves on the road to estimate how stress levels will evolve in the next minute. Different machine learning techniques are evaluated and the results for both intra and inter-city driving and for both intra and inter driver data are presented. We have used data from four different drivers with three different car models and a motorbike and more than 220 test drives. Results show that upcoming stress levels can be accurately predicted for a single user (correlation r = 0.99 and classification accuracy 97.5%) but prediction for different users is more limited (correlation r = 0.92 and classification accuracy 46.9%).
No takes yet. Share an insight, caveat, or question.
Muñoz-Organero et al. (2016) conducted an observational in Stress while driving (n=4). Machine learning models using current stress, driving behavior, and road shape was evaluated on Prediction of upcoming stress levels in the next minute (r = 0.99 (single user), r = 0.92 (different users)). Machine learning models accurately predicted upcoming stress levels for a single user (accuracy 97.5%, r=0.99), but performance was limited when predicting across different users (accuracy 46.9%).
Synapse has enriched 10 related papers. Consider them for comparative context: