Geographic Probability Algorithms with Security Force Applications

Every day, government executives, police officials, and military leaders must decide how to most efficiently and effectively employ their limited resources in an effort to secure the large and diverse populations they are charged to protect. Increasingly, these leaders rely on the analytic tools provided by the discipline of crime analysis. One of the most important tools in the discipline of crime analysis is the predictive hot-spot map, which is used to make tactical level decisions about the employment of resources. This dissertation develops methodological approaches for exploiting these predictive crime maps to improve the crime forecasts, geographic districting plans, intelligence assessments, and targeting plans that support military and police decision makers.

This research provides four multidisciplinary contributions. First, this dissertation provides a new method for forecasting noisy geographic time series that provides statistically significant performance improvements over the most-used forecasting methods while dramatically reducing modeling workload so long as several modeling assumptions are satisfied. Second, this new forecasting method is supported by the development of a statistical motivation that explains why weighted aggregate forecasts provide better forecasting performance for disaggregated event count time series than forecasts made using the observations from the many disaggregated event count time series themselves. Third, this dissertation documents a new method for geographically mapping the region where spatial choice behavior by one entity or group will dominate spatial choice behavior by all other considered groups. Finally, this dissertation documents the development of a new approach for Journey to Crime (JTC) analysis that adds to the existing literature by providing the ability to simultaneously model the effect of many environmental factors on the spatial choice behavior of the modeled agents (plants, animals, or criminals) while incorporating the distance-decay modeling used by existing JTC methods.

This dissertation demonstrates the practical application of these research contributions in four case studies. First, a new geographic forecasting method, Geographic Probability Forecasting (GPF), is applied to the problem of forecasting weekly burglary counts over a five-year period in Pittsburgh, Pennsylvania. The GPF method links the tactical and operational levels of planning, reduces modeling workload, and significantly improves forecasting performance for this problem. Second, the GPF method is leveraged to produce planning maps for Albemarle County, Virginia, that facilitate the development and evaluation of the districting plans that are used to define geographic areas of responsibility for patrolling units. Third, previous work in Criminal Site Selection (CSS) modeling is extended to develop a Sphere of Influence (SOI) analysis, improving the intelligence assessments for criminal gangs in Santa Ana, California. Finally, CSS models are leveraged to develop a new JTC analysis technique that outperforms the current best JTC method for predicting the geographic anchor points of criminal gangs in Santa Ana.