Boosting Algorithms for the Accident Severity Classification DOI

Islam Babaev,

Igor Mozolin,

Divya Garikapati

et al.

SAE International Journal of Connected and Automated Vehicles, Journal Year: 2024, Volume and Issue: 8(4)

Published: Oct. 17, 2024

<div><i>Background:</i> Road accident severity estimation is a critical aspect of road safety analysis and traffic management. Accurate contributes to the formulation effective policies. Knowledge potential consequences certain behaviors or conditions can contribute safer driving practices. Identifying patterns high-severity accidents allows for targeted improvements in terms overall safety. <i>Objective:</i> This study focuses on analyzing by utilizing real data, i.e., US open database called “CRSS.” It employs advanced machine learning models such as boosting algorithms LGBM, XGBoost, CatBoost predict classification based various parameters. The also aims providing predictive insights stakeholders, functional engineering community, policymakers using KABCO systems. article includes sections covering theoretical methodology, data analysis, model development, evaluation, performance metrics, implications improving measures comparing different CRSS dataset. identify most algorithm integrate into our product line near future, enabling accurate prediction both occurrence. <i>Results Conclusions:</i> addresses challenges evaluating metrics classes within unbalanced datasets, emphasizing impact dominant like Class O (O = no apparent injury) accuracy. investigation reveals limitations conservatism associated with imbalanced models, hinting at ceiling their around 80%. Comparative algorithms, including CatBoost, demonstrates comparable even case applying KNN pre-processing, especially accuracy, <i>F</i><sub>1</sub>-score, ROC-AUC, PR-AUC all classes. XGBoost did not show any significant improvement compared without algorithm. CM upper triangle, applied an study. Future work directions involve extending application other diverse exploring capabilities deep neural networks, refining dataset preparation accuracy improvement, creating unified tools hazard risk assessment.</div>

Language: Английский

Visual Recognition Distance Prediction and Energy-Efficient Brightness Optimization Strategies for Active Luminous Guide Signs DOI
Yichang Shao, Zhirui Ye,

Renhao Hu

et al.

Transportation Research Record Journal of the Transportation Research Board, Journal Year: 2025, Volume and Issue: unknown

Published: April 27, 2025

In response to increasing urbanization and vehicle ownership, road traffic safety has become a critical area of study, especially under conditions poor visibility at night or in adverse weather. Traditional retroreflective signs often fail provide sufficient visibility, the risk accidents. This research focuses on active luminous guide signs, novel management technology that seen widespread adoption across various regions. Our paper explores visual recognition distance these by analyzing aspects brightness level, speed, ambient light conditions. Real experiments were conducted study how those factors affect distance. The utilizes genetic algorithm–backpropagation (GA-BP) neural network models simulate employs multi-objective algorithm optimize both energy consumption based non-dominated sorting algorithm-II (NSGA-II) technique for order preference similarity ideal solution (TOPSIS). By delineating optimal parameters different lighting speeds, this offers strategic insights into contributing significantly infrastructure digital transformation.

Language: Английский

Citations

0
BABE: Backdoor attack with bokeh effects via latent separation suppression DOI
Junjian Li, Honglong Chen, Yudong Gao

et al.

Engineering Applications of Artificial Intelligence, Journal Year: 2024, Volume and Issue: 138, P. 109462 - 109462

Published: Oct. 22, 2024

Citations

0
Boosting Algorithms for the Accident Severity Classification DOI

Islam Babaev,

Igor Mozolin,

Divya Garikapati

et al.

SAE International Journal of Connected and Automated Vehicles, Journal Year: 2024, Volume and Issue: 8(4)

Published: Oct. 17, 2024

<div><i>Background:</i> Road accident severity estimation is a critical aspect of road safety analysis and traffic management. Accurate contributes to the formulation effective policies. Knowledge potential consequences certain behaviors or conditions can contribute safer driving practices. Identifying patterns high-severity accidents allows for targeted improvements in terms overall safety. <i>Objective:</i> This study focuses on analyzing by utilizing real data, i.e., US open database called “CRSS.” It employs advanced machine learning models such as boosting algorithms LGBM, XGBoost, CatBoost predict classification based various parameters. The also aims providing predictive insights stakeholders, functional engineering community, policymakers using KABCO systems. article includes sections covering theoretical methodology, data analysis, model development, evaluation, performance metrics, implications improving measures comparing different CRSS dataset. identify most algorithm integrate into our product line near future, enabling accurate prediction both occurrence. <i>Results Conclusions:</i> addresses challenges evaluating metrics classes within unbalanced datasets, emphasizing impact dominant like Class O (O = no apparent injury) accuracy. investigation reveals limitations conservatism associated with imbalanced models, hinting at ceiling their around 80%. Comparative algorithms, including CatBoost, demonstrates comparable even case applying KNN pre-processing, especially accuracy, <i>F</i><sub>1</sub>-score, ROC-AUC, PR-AUC all classes. XGBoost did not show any significant improvement compared without algorithm. CM upper triangle, applied an study. Future work directions involve extending application other diverse exploring capabilities deep neural networks, refining dataset preparation accuracy improvement, creating unified tools hazard risk assessment.</div>

Language: Английский

Citations

0