도시 기능구역별 운전자의 가시적 경관이 자동차 교통사고에 미치는 영향 분석 : 해석가능한 기계학습과 음이항 회귀모형의 혼합적 접근을 중심으로 = Analysis of the Impact of Driver’s Visual Landscape on Traffic Accidents in Urban Functional Zones : A Hybrid Approach Combining Interpretable Machine Learning and Negative Binomial Regression

다국어 초록 (Multilingual Abstract)

Urban functional zones exhibit distinct patterns of traffic accidents, reflecting variations in spatial configuration and visual landscapes. This study investigates how drivers’ visual landscape elements influence automobile traffic accidents in Seoul. Based on 551,804 Point of Interest (POI) records obtained from Kakao Maps, the city was categorized into four functional zones using Word2Vec embeddings and K-means clustering: high-density residential and educational areas; urban mixed-use activity areas; transit and leisure mixed-use areas; and low-density residential and local living areas. Visual landscape indices, including the Green View Index (GVI), Color Entropy Index (CEI), and Visual Obstruction Index (VOI), were derived from 95,842 Naver Street View images through semantic segmentation and color analysis. XGBoost with SHAP values was used to identify influential variables, followed by negative binomial regression for statistical validation. The results indicate that CEI reduces accidents in high-density residential and educational areas; GVI reduces accidents in urban mixed-use activity areas; VOI increases accidents in transit and leisure mixed-use areas; and GVI increases accidents in low-density residential and local living areas. These findings suggest that the effects of visual landscape elements differ across functional zones, underscoring the need for tailored urban design and traffic safety strategies.