Risk Assessment Score Based on Simulated Driving Session
Authors:
Mojca Komavec, Boštjan Kaluža, Kristina Stojmenova, Jaka Sodnik
Keywords:
driving simulator, driver evaluation, risk assessment, neural networks
Abstract:
Driving simulators can be used for training and evaluation of driving skills. However, for an effective training process, a scoring system with relevant feedback should be applied. Such scoring system can be then used for providing feedback on lacking skills for customized training or for estimation of potential risky driving behaviour. In this paper, we present a new approach for evaluation of the driver, which is based on data obtained through a simulated driving session with variety of challenging roads and traffic conditions and data acquired from biometrical sensors. First, we present a rule-based scoring model for driver evaluation, which predicts the likelihood of a driver being involved in risky behaviour. In the second part we present a process of recalibration of weights of the scoring model using neural networks and Naïve Bayes algorithm. The results showed that neural networks outperformed all of other tested models, as it achieved 23% better results in terms of classification accuracy, and 7% better results in terms of precision when compared to the initial rule-based model. However, by using machine learning approach, the interpretation of final score and weights is very limited and valuable feedback on lacking skills is lost. We evaluate such scoring systems as appropriate for risk assessment, while rule-based models prove to be more appropriate for driver training.
Cite this article
Komavec M.; Kaluža B.; Stojmenova K. and Sodnik J. Risk Assessment Score Based on Simulated Driving Session In: Proceedings of the Driving Simulation Conference 2019 Europe VR, Driving Simulation Association, Strasbourg, France, 2019, pp. 67-74
@inproceedings{Komavec2019, title = {Risk Assessment Score Based on Simulated Driving Session}, author = {Mojca Komavec and Boštjan Kaluža and Kristina Stojmenova and Jaka Sodnik}, editor = {Andras Kemeny and Florent Colombet and Frédéric Merienne and Stéphane Espié}, isbn = {978-2-85782-749-8}, year = {2019}, date = {2019-09-04}, booktitle = {Proceedings of the Driving Simulation Conference 2019 Europe VR}, pages = {67-74}, address = {Strasbourg, France}, organization = {Driving Simulation Association}, abstract = {Driving simulators can be used for training and evaluation of driving skills. However, for an effective training process, a scoring system with relevant feedback should be applied. Such scoring system can be then used for providing feedback on lacking skills for customized training or for estimation of potential risky driving behaviour. In this paper, we present a new approach for evaluation of the driver, which is based on data obtained through a simulated driving session with variety of challenging roads and traffic conditions and data acquired from biometrical sensors. First, we present a rule-based scoring model for driver evaluation, which predicts the likelihood of a driver being involved in risky behaviour. In the second part we present a process of recalibration of weights of the scoring model using neural networks and Naïve Bayes algorithm. The results showed that neural networks outperformed all of other tested models, as it achieved 23% better results in terms of classification accuracy, and 7% better results in terms of precision when compared to the initial rule-based model. However, by using machine learning approach, the interpretation of final score and weights is very limited and valuable feedback on lacking skills is lost. We evaluate such scoring systems as appropriate for risk assessment, while rule-based models prove to be more appropriate for driver training.}, keywords = {driver evaluation, driving simulator, neural networks, risk assessment}, }
TY - CONF TI - Risk Assessment Score Based on Simulated Driving Session AU - Komavec, Mojca AU - Kaluža, Boštjan AU - Stojmenova, Kristina AU - Sodnik, Jaka C1 - Strasbourg, France C3 - Proceedings of the Driving Simulation Conference 2019 Europe VR DA - 2019/09/04 PY - 2019 SP - 67 EP - 74 LA - en-US PB - Driving Simulation Association SN - 978-2-85782-749-8 L2 - https://proceedings.driving-simulation.org/proceeding/dsc-2019/risk-assessment-score-based-on-simulated-driving-session ER -
A Driving Simulation Association membership is required in order to download the papers.