This study presents a hybrid framework that integrates machine learning and network science methods to analyze urban crime dynamics from both structural and spatial perspectives. Using incident data from the Boston Police Department, the authors investigate two complementary dimensions of crime behavior. First, an offense co-occurrence network is constructed using multi-offense incidents, and a co-occur offenses index is derived from eigenvector centrality to identify structurally influential offense types. Second, a PCA-based scoring mechanism is applied to quantify long-term crime intensity across spatial subregions, producing continuous hot spot scores that reflect both spatial and temporal consistency. The results reveal a highly uneven distribution of crime patterns. A small number of offenses and locations exhibit disproportionately high structural centrality and crime persistence, while the majority remain secondary. These findings support targeted enforcement strategies and contribute to a deeper understanding of crime concentration. The proposed framework offers a scalable and interpretable tool for crime pattern analysis and decision-making in public safety planning.