Related Experiment Video
Updated: Jul 29, 2026
16:14
Trajectory Data Analyses for Pedestrian Space-time Activity Study
Published on: February 25, 2013
Key feature selection and risk prediction for lane-changing behaviors based on vehicles' trajectory data
Tianyi Chen1, Xiupeng Shi1, Yiik Diew Wong1
1School of Civil and Environmental Engineering, Nanyang Technological University, 639798, Singapore.
Accident; Analysis and Prevention
|June 1, 2019
Summary
This study introduces a framework for predicting risky lane-changing (LC) behavior using vehicle trajectory data. It identifies key features and uses the Crash Potential Index (CPI) to assess risk, improving highway traffic safety.
Area of Science:
- Traffic Safety Engineering
- Machine Learning Applications in Transportation
- Behavioral Analysis of Road Users
Background:
- Risky lane-changing (LC) behavior significantly compromises road traffic safety.
- Existing research lacks a comprehensive framework for identifying key features and predicting LC risk on highways.
- Vehicle trajectory data offers a rich source for analyzing driving behaviors and associated risks.
Purpose of the Study:
- To develop and validate a research framework for key feature selection and risk prediction of vehicle LC behavior on highways.
- To identify critical features influencing LC risk using advanced analytical techniques.
- To evaluate and propose optimal methods for handling imbalanced datasets in LC risk prediction.
Main Methods:
- Extraction of candidate features from vehicle trajectory datasets (NGSIM).
- Application of Fault Tree Analysis and k-Means clustering to determine LC risk levels via the Crash Potential Index (CPI).
- Utilized Random Forest (RF) classifier for feature selection and risk prediction, and evaluated resampling methods like SMOTETomek.
Main Results:
- Sensitivity analysis of CPI revealed that following vehicles in the original lane are safest, while those in the target lane are riskiest during LC events.
- SMOTETomek demonstrated superior performance for resampling imbalanced LC risk datasets, minimizing overfitting and enhancing prediction accuracy for the RF classifier.
- Key features influencing LC risk include individual behaviors of the LC car and surrounding vehicles, inter-car interactions, and particularly, acceleration interactions in the target lane.
Conclusions:
- The proposed framework effectively identifies key features and predicts LC risk, contributing to enhanced traffic safety.
- The study highlights the importance of specific vehicle interactions and accelerations in assessing lane-changing risks.
- Optimized data resampling techniques are crucial for building accurate predictive models for risky driving behaviors.
