|
home / centre for automotive safety research / Publications / List / Details Publication Details| Title | Predicting Driver Sleep history with Machine Learning - A Simulator-Based Study | | Authors | Tuckwell GA | | Year | 2025 | | Type | Unknown | | Abstract | Fatigued driving is a major road safety concern and is a contributing factor to approximately 20% of all road traffic accident worldwide. Identifying a driver’s recent sleep duration may help to detect at-risk drivers before they make an error. Participants (n = 84, 23.6 ± 4.5 years) completed a 7-day laboratory study and were randomly allocated to a sleep condition: 9h or 5h sleep opportunity. Each participant completed two 20-min simulated drives (n = 795) at 8:10h and 17:30h each day. A random forest model was implemented using the output of the driving simulator to classify prior sleep duration (9h or 5h). Accuracy was determined using 3-fold cross validation, with the highest accuracy produced by the model of 65% (F-score = 0.64). High importance features identified by the random forest for classification included lane position and speed. Preliminary results suggest that this approach is suitable for classifying prior sleep duration with further model hyperparameter tuning required to increase model performance. | | Conference Name | 2025 Australasian Road Safety Conference | | Conference Abbreviation | ARSC25 | | Conference Location | Perth, Western Australia | | Conference Date | 20-23 October 2025 |
| Reference | | Tuckwell GA (2025). Predicting Driver Sleep history with Machine Learning - A Simulator-Based Study. 2025 Australasian Road Safety Conference, Perth, Western Australia, 20-23 October 2025. [PRESENTED ABSTRACT] |
|
