Research on Performance Risk Assessment and Explainability Modeling of AEB Systems Based on 24th Edition C-Ncap Test Scenario Data

2026-99-2420

To be published on 11/20/2026

Authors
Abstract
Content
Regarding the issue in the 2024 version of C-NCAP’s automatic emergency braking (AEB) test where the scene score rates are highly concentrated at the maximum score, and traditional continuous regression modeling is difficult to effectively depict system risks, this paper transforms the AEB performance assessment problem into a rare failure risk prediction problem, and constructs an AEB failure risk assessment method based on LightGBM and SHAP. Based on 464 test samples, an explanatory variable system is established around operating condition parameters, vehicle configuration parameters, and process temporal sequence parameters, and samples with scene score rates lower than 1 are defined as failure samples. Firstly, the original data undergoes outlier removal, missing value handling, categorical variable encoding, and continuous variable standardization; secondly, a robust LightGBM classification model is constructed to predict AEB deduction/ failure risk; finally, the SHAP method is used to conduct interpretability analysis of the model to identify the key factors affecting AEB system performance. The results show that the established robust model achieves good recognition effects on the test set, with an AUC of 0.871, PR-AUC of 0.527, and F1 of 0.667. The feature importance and SHAP analysis results show that the number of sensors, relative speed, alarm-time TTC, braking-time TTC, overlap rate, and maximum braking deceleration are the key variables affecting AEB failure risk. Among them, higher sensor redundancy and earlier alarm and braking trigger times can significantly reduce system failure risk, while high relative speed, complex collision geometric relationship, and braking intervention lag will significantly increase the failure probability. The proposed method can effectively identify AEB system performance risks under the condition of highly concentrated score rates and rare failure samples, providing a data-driven analysis tool for AEB calibration optimization, sensor configuration design, and active safety performance improvement.
Meta TagsDetails
Citation
Yuan, Y., Li, Y., Zhang, C., and Li, Z., "Research on Performance Risk Assessment and Explainability Modeling of AEB Systems Based on 24th Edition C-Ncap Test Scenario Data," 2026 2nd International Conference on Intelligent Transportation and Future Mobility (ITFM2026), Xi'An, China, March 27, 2026, .
Additional Details
Publisher
Published
To be published on Nov 20, 2026
Product Code
2026-99-2420
Content Type
Technical Paper
Language
English