Park, Peter2019-07-022019-07-022018-12-172019-07-02http://hdl.handle.net/10315/36249This research contributes to developing macro-level crime and collision prediction models using a new method designed to handle the problem of spatial dependency and over-dispersion in zonal data. A geographically weighted Poisson regression (GWPR) model and geographically weighted negative binomial regression (GWNBR) model were used for crime and collision prediction. Five years (2009-2013) of crime, collision, traffic, socio-demographic, road inventory, and land use data for Regina, Saskatchewan, Canada were used. The need for geographically weighted models became clear when Moran's I local indicator test showed statistically significant levels of spatial dependency. A bandwidth is a required input for geographically weighted regression models. This research tested two bandwidths: 1) fixed Gaussian and 2) adaptive bi-square bandwidth and investigated which was better suited to the study's database. Three crime models were developed: violent, non-violent and total crimes. Three collision models were developed: fatal-injury, property damage only and total collisions. The models were evaluated using seven goodness of fit (GOF) tests: 1) Akaike Information Criterion, 2) Bayesian Information Criteria, 3) Mean Square Error, 4) Mean Square Prediction Error, 5) Mean Prediction Bias, and 6) Mean Absolute Deviation. As the seven GOF tests did not produce consistent results, the cumulative residual (CURE) plot was explored. The CURE plots showed that the GWPR and GWNBR model using fixed Gaussian bandwidth was the better approach for predicting zonal level crimes and collisions in Regina. The GWNBR model has the important advantage that can be used with the empirical Bayes technique to further enhance prediction accuracy. The GWNBR crime and collision prediction models were used to identify crime and collision hotzones for simultaneous crime and collision reduction in Regina. The research used total collision and total crimes to demonstrate the determination of priority zones for focused law enforcement in Regina. Four enforcement priority zones were identified. These zones cover only 1.4% of the Citys area but account for 10.9% of total crimes and 5.8% of total collisions. The research advances knowledge by examining hotzones at a macro-level and suggesting zones where enforcement and planning for enforcement are likely to be most effective and efficient.enAuthor owns copyright, except where explicitly noted. Please contact the author directly with licensing requests.Transportation PlanningElectronic Thesis or Dissertation2019-07-02Crime and Collision DataMacro-level ModellingSpatial DependencySpatial AnalysisGeographically Weighted Negative Binomial RegressionGeographically Weighted Poisson RegressionHotzone IdentificationLaw Enforcement