서울시 고령 운전자의 중상 및 사망사고와 도로 환경의 연관성 분석: 기계학습을 활용한 비선형성과 상호작용 효과를 중심으로

초록

This study analyzes traffic accident data from the TAAS (2017-2019) to examine non-linear relationships and interaction effects between road environmental factors and fatal accidents involving older adult drivers in Seoul. Machine learning-based tree ensemble algorithms and the SHAP method were used to provide insights integrating theoretical perspectives. The main findings are as follows: First, the XGBoost model performed best in predicting accidents for both older adult drivers and non-older adult drivers, with significant improvement observed for non-older adult drivers using the SMOTE technique. Second, for older adult drivers, road network characteristics and complex environments (intersection count, commercial facility density) were key factors, while non-older adult drivers showed more flexibility, reducing the importance of these factors. Green space proportion was found to lower accident probability for older adult drivers. Third, non-linear patterns were observed in variables such as Closeness, commercial facility density, and openness, with accident probability rising sharply beyond certain thresholds. Finally, interaction effects based on Attention Restoration Theory indicated that in areas with high commercial facility density or enclosed environments, an optimal green space ratio reduced accident probability. These results suggest that green spaces can enhance attention and reduce accidents for older adult drivers in complex environments. This study informs road environment design and traffic policy development to reduce older adult drivers accidents, highlighting the need for tailored safety strategies based on non-linear patterns and interactions.

키워드