A Machine Learning Approach to Analyzing Crime Concentration: The Case of New York City

Abstract

Building upon prior work, we propose an alternative way to look at the pattern of spatial crime concentration and temporal stability of it. We first identify a high-crime cluster using the sample block groups in New York City by employing a k-means clustering method. We then examine the temporal stability of the high-crime cluster over time. We also longitudinally assess how our high-crime cluster classification is associated with the actual amount of crime while accounting for the measures of social and physical environments. We observed that about 6–12% of total areas are identified to be in the high-crime cluster. We also found that block groups identified to be high-crime cluster in one year are more likely to be that way in the next year. We hope future research may consider using data-driven approaches to expand understanding of spatial and temporal crime patterns.