개별 차량 주행행태 자료를 이용한 앙상블 학습전략 기반 도로 위험 구간 식별 기법

Identification of Crash-Prone Road Segments Based on an Ensemble Machine Learning Strategy : Use of Driving Behavior Data
  • 지정훈; 
  • 구예서 ; 
  • 조영; 
  • 오철
Citations

SCOPUS

2

초록

Various road safety evaluation studies have been conducted for traffic safety management. To date, traffic safety measures have been primarily focused on reactive safety management using macroscopically aggregated traffic data collected by infrastructure-based detection systems. However, the need for proactive safety management has been raised as individual vehicle driving behavior data is now readily available. Detection of crash prone road segments allows for realizing more proactive safety management to prevent traffic crashes. This study applied statistical techniques and machine learning models based on individual vehicle data capable of analyzing driving behavior. It also aims to implement a more accurate road hazard identification model using an ensemble learning strategy. Twenty-two driving safety indicators were derived to characterize vehicle driving behavior from inter-vehicular interaction, longitudinal, and lateral perspectives, based on the data collected from PVD, which provides individual vehicle driving behavior data.Ten of these indicators, which showed significant correlation with traffic accident data, were selected through statistical correlation analysis. The average number of accidents in the analysis section of traffic accident data was used to categorize road sections into hazardous and normal categories, serving as the target variable. The ten statistically significant driving safety indicators were used as explanatory variables. Classification models such as SVM, ANN, and KNN were performed. Ensemble learning, known for enhancing robustness compared to single models, was employed by combining these three models to derive the final model. The accuracy of the model by ensemble learning strategy was 90.8%, showing an improvement of up to 4.8% and at least 2.7% compared to single models. This indicates that ensemble learning strategies can enable more accurate implementation of road hazard identification models for proactive traffic safety management. The methodology of this study is expected to serve as foundational data for implementing models for real-time crash prone road segments identification in environments where real-time data collection and utilization are feasible.

키워드

도로 위험 구간 식별 모델; 주행 안전성 평가지표; 머신러닝; 앙상블 학습전략; 선제적 교통안전 관리; crash prone road segments detection model; driving safety indicator; machine learning; ensemble learner; proactive traffic safety management
제목
개별 차량 주행행태 자료를 이용한 앙상블 학습전략 기반 도로 위험 구간 식별 기법
제목 (타언어)
Identification of Crash-Prone Road Segments Based on an Ensemble Machine Learning Strategy : Use of Driving Behavior Data
저자
지정훈; 구예서 ; 조영; 오철
DOI
10.7470/jkst.2023.41.6.739
발행일
2023-12
유형
정기학술지(Article(Perspective Article포함))
저널명
대한교통학회지
권
41
호
6
페이지
739 ~ 752