Manhattan, a central US hub for economics and entertainment, consistently battles high crime rates. Despite existing research examining correlations between overall crime, spatial and sociodemographic factors, or urban architecture's influence on specific crimes, there's a dearth of empirical studies integrating street-level urban features and sociodemographic contexts in sexual crime analysis. This study employs computer vision, machine learning, and big data to investigate associations between Manhattan's urban environment and street-level sex crimes. Two initial Ordinary Least Squares (OLS) models examine the distribution of these crimes from macro-urban and micro-environmental perspectives. A Geographically Weighted Regression (GWR) model further explores local sex crime correlations with different spatial-scale variables. The results suggest a model incorporating multi-dimensional microenvironmental characteristics more effectively explains the incidence of sexual crimes.
Keywords: Sexual Violence, Semantic Segmentation, Space Syntax, Least Square Regression (OLS), Geographic Weighted Regression (GWR)