Localized Crime Prediction Methods

The convergence of public data and statistical modeling has created opportunities for public safety officials to prioritize the deployment of scarce resources on the basis of predicted crime patterns. Current crime prediction methods are trained using observed crime and information describing various criminogenic factors. Researchers have favored global models (e.g., of entire cities) due to a lack of observations at finer resolutions (e.g., ZIP codes). These global models and their assumptions are at odds with evidence that the relationship between crime and criminogenic factors is not homogeneous across space. In response to this gap, this dissertation presents a framework for building localized crime prediction models. The proposed models achieve localization using three approaches: 1) quantifying micro-level daily routine features from social media; 2) building area-specific models at finer resolutions (e.g, neighborhood), and 3) proposing a new way for estimating historical crime density at the local level. Experimental results on real crime data from Chicago, Illinois indicate predictive advantages over multiple state-of-the-art global models. Furthermore, this dissertation includes a comprehensive performance analysis of existing evaluation metrics and a human factor study on the effectiveness of visualization techniques for making decisions that are informed by forecasts. The dissertation follows a holistic approach for crime prediction modeling such that it is not only important to build high quality models but also properly evaluate them and effectively visualize their outcomes. This research helps security agents to better allocate their resources and people to better manage crime risks, which ultimately improve public safety.