Development of Advanced Computational and Geographic Visualization Methods for Geospatial and Temporal Strategic Risk Assessment of Crime

As submitted by the proposer: There are several known attractors of crime, like drug markets, and bus stops (Block and Block, 1995; Brantingham and Brantingham, 1995; Weisburd and Green, 1995; Hart and Miethe, 2014), and land use concentrations (e.g. bars, motels and public housing).

Current cluster visualization methods (e.g. hot-spot mapping and kernel density estimation) are limited in their approach to revealing non-obvious attractors of crime across space and time.

The purpose of this study is to develop computational, multi-dimension geospatial and temporal attribute data models to discover non-obvious crime attractors in institutionalized, high-density cluster locations. Such models are expected to be a significant improvement over current models that tend to examine a few variables in space-time phenomenon.

Crime data (2004-2014) collected from Atlanta, Chicago and Dallas Police Departments will be combined with non-police data (e.g. census and land use) to answer the following questions: (1)What spatial-temporal patterns do the study areas share in the selected crime categories? If no commonalities exist, then what attributes contribute to their regional differences? (2)

For locations with institutionalized, high-density clustering of crime, what is the efficacy level for the use of self organizing maps [SOMs] in non-obvious attractor discovery in both geographic and attribute space? (3) If shared attribute and spatial-temporal patterns do exist, what type of strategic forecasting model may be implemented to assess neighborhood vulnerability that will also perform well in a different region? The analyses will occur at two micro-scale resolutions: 1) street segments, and 2) census blocks. To address the first question, long-term, high-density clusters will be identified in each city using exploratory regression and statistical density methods.

The remaining research questions will be addressed using a combination of R programming packages and Q-GIS to test the following hypotheses: (1) Regardless of regional location, it is possible for multiple jurisdictions to experience very similar crime event patterns. (2) Attribute space analysis will reveal non-obvious relationships which may not necessarily lend themselves to geographic space; however, those hidden relationship patterns will have a significant measurable effect on crime and how those events cluster over time. One city will be selected for training and validation of the models. The remaining cities will serve as test locations to evaluate the accuracy of the models on different regions.

Research findings will be reported in the dissertation, peer-reviewed publications, a summary report and conference presentations. Products expected include computational models for R packages.

This project contains a research and/or development component, as defined in the applicable law, and complies with Part 200 Uniform Requirements – 2 CFR 200.210(a) (14). nca/ncf