논문 상세보기
설명 가능한 인공지능(XAI)을 활용한 회전교차로 기하구조의 안전성 임계값 규명 - 설계 용량 지수(DCI)와 EPDO Rate를 중심으로 KCI 등재
Identification of Safety Efficiency Thresholds in Roundabout Geometrics using Explainable Artificial Intelligence - Focusing on Design Capacity Index (DCI) and Equivalent Property Damage Only (EPDO) Rate
pp.49-59
As domestic traffic policies have shifted from vehicle-centric approaches to a ‘Safe Speed’ paradigm, the installation of roundabouts has surged. However, existing studies based on linear statistical models have failed to identify the complex non-linear interactions between geometric features and accident severity, limiting their ability to provide concrete design thresholds. To overcome the lack of traffic volume data, this study developed a geometry-based Design Capacity Index (DCI) and proposed a new analytical framework using the Equivalent Property Damage Only (EPDO) rate per unit capacity as the dependent variable. Utilizing a dataset of nationwide roundabouts (2007–2020), a grid search-optimized eXtreme Gradient Boosting (XGBoost) model and SHAP analysis were applied, achieving a 40.5 % performance improvement over linear baselines. The results revealed that circulatory roadway width was a dominant factor; contrary to the 'Road Diet' theory, ensuring 'Geometric Sufficiency' (wider lanes) proved more effective for safety in medium-to-large roundabouts. Furthermore, a 'Broad Optimal Zone' was identified within an inscribed circle diameter (ICD) of 35–70 m, while a 'Paradox of Scale' emerged beyond 70 m where safety benefits plateaued. Additionally, raised crosswalks served as essential offset measures, consistently reducing accident costs regardless of the intersection size. Based on these findings, this study provides empirical evidence for revising design guidelines to prioritize the 35–70 m ICD range and advocates for the mandatory installation of physical calming measures in oversized roundabouts.
1. 서론
2. 이론적 고찰
2.1. 분석 방법론
2.2. 선행연구 고찰
2.3. 선행연구와의 차별성
3. 연구 설계
3.1. 분석 프레임워크 및 연구 가설
3.2. 데이터 구축 및 전처리
3.3. 주요변수 산정 및 정의
3.4. 분석 모형의 설계 및 적용
4. 분석 결과
4.1. 모형의 적합도
4.2. 주요 영향 요인 중요도 분석
4.3. 기하구조의 한계 효과 및 비선형 임계값 분석
5. 결론
5.1. 결론 및 시사점
5.2. 연구의 한계 및 향후 연구과제
REFERENCES
