Interaction Effects Among Risk Domains in Commercial Fleet Crash Probability Estimation
Annotation
A weighted sum of five risk-domain scores rates a window with two severely degraded domains below a window with five moderately elevated ones, so windows in which driver, vehicle, route and duty schedule fail together are underrated. The framework introduced in this study adds ten pairwise domain products to a five-domain composite, estimates the probability of a safety-critical event within 30 minutes from the augmented vector, calibrates it by isotonic regression and ranks interventions by counterfactual risk reduction. In 20 simulated fleets with about 6,900 training events, a logistic estimator with explicit products lowered the expected calibration error in joint-exposure windows (two or more domain scores at or above 60) from 0.041 to 0.024 and the intervention-ranking regret from 0.20 to 0.07 percentage points relative to an additive logistic estimator; ROC-AUC stayed at 0.820 for both. The calibration gain appears from about 700 training events and the ranking gain from about 2,300; below roughly 250 events the additive estimator ranks better. A boosted ensemble given the same products gained nothing, and calibrated boosted trees trailed the additive logistic estimator on Brier score at every training size from 81 to 6,900 events. A 12-week pilot on 15 tractors over 345,000 km reported a calibrated ROC-AUC of 0.815, a Brier score of 0.082 and 1.4 false alerts per 100 driving hours.
Keywords
References:
Chen, T., & Guestrin, C. (2016). XGBoost: A scalable tree boosting system. In Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (pp. 785-794). ACM. https://doi.org/10.1145/2939672.2939785
Dingus, T. A., Guo, F., Lee, S., Antin, J. F., Perez, M., Buchanan-King, M., & Hankey, J. (2016). Driver crash risk factors and prevalence evaluation using naturalistic driving data. Proceedings of the National Academy of Sciences, 113(10), 2636-2641. https://doi.org/10.1073/pnas.1513271113
Guo, C., Pleiss, G., Sun, Y., & Weinberger, K. Q. (2017). On calibration of modern neural networks. In Proceedings of the 34th International Conference on Machine Learning (Vol. 70, pp. 1321-1330). PMLR.
Leveson, N. (2004). A new accident model for engineering safer systems. Safety Science, 42(4), 237-270. https://doi.org/10.1016/S0925-7535(03)00047-X
Mannering, F., Bhat, C. R., Shankar, V., & Abdel-Aty, M. (2020). Big data, traditional data and the tradeoffs between prediction and causality in highway-safety analysis. Analytic Methods in Accident Research, 25, Article 100113. https://doi.org/10.1016/j.amar.2020.100113
Niculescu-Mizil, A., & Caruana, R. (2005). Predicting good probabilities with supervised learning. In Proceedings of the 22nd International Conference on Machine Learning (pp. 625-632). ACM. https://doi.org/10.1145/1102351.1102430
Parsa, A. B., Movahedi, A., Taghipour, H., Derrible, S., & Mohammadian, A. (2020). Toward safer highways, application of XGBoost and SHAP for real-time accident detection and feature analysis. Accident Analysis & Prevention, 136, Article 105405. https://doi.org/10.1016/j.aap.2019.105405
Rudin, C. (2019). Stop explaining black box machine learning models for high stakes decisions and use interpretable models instead. Nature Machine Intelligence, 1(5), 206-215. https://doi.org/10.1038/s42256-019-0048-x
VanderWeele, T. J., & Knol, M. J. (2014). A tutorial on interaction. Epidemiologic Methods, 3(1), 33-72. https://doi.org/10.1515/em-2013-0005
Other articles of the issue
