GIS5100 Module 1: Crime Analysis

    This week’s lab in Applications in GIS allowed us to practice different hotspot mapping techniques – part of lab requested we use provided data to determine hotspot maps of Chicago, Illinois for 2017 and be able to determine the accuracy of each map to predict the area of homicides in 2018. We created (3) hotspot maps with (3) different methods; Grid-based Thematic, Kernel Density, and Local Moran’s I.

 

Grid-based Thematic Map (upper left corner, yellow), Kernel Density Map (upper right corner, purple), and Local Moran's I Map (bottom center, pink). Each map depicts hotspots of homicides in Chicago, Illinois in 2017.

Grid-based Thematic:

A Spatial Join was done by joining the features classes of the Chicago grid and 2017 total homicides (Join Operation: One to One, Match Option: Intersect). A Select By Attributes was used to select the grids that had the homicide count greater than zero and exporting that selected data into a new feature class. In the newly exported feature class, the top 20% of grid cells with the highest count were determined – to do this, I used the total number of counts (311) and divided by five to determine the top quintile- in this instance, the top quintile is 62 (311/5 = 62.2 = 62). I then selected the top 62 records in the feature class and exported this as a feature class: hom_top20per. This feature class was then dissolved using Dissolve.

 

 

 

Kernel Density

 

The Kernel Density was determined using the Kernel Density tool with the total homicides of 2017 data with the Chicago Boundary data as the barrier feature.

 

Under Symbology -> Statistics: The mean in this feature is (2.88), and maximum value (38.92). These figures were used in determining the two break values. Below are the two break values:

·       2.88 * 3 = 8.64

·       38.92 (max value)

I used the two break values above in when using the Reclassify tool as the Reclassification values, then used Raster to Polygon tool and Select by Attributes to only select the grid code value of (2). The grid value (2) are all the values that were between my two break values, 8.64 - 38.92.

 

Local Moran’s I

A Spatial Join was done using the Census Tracts and total homicides of 2017 data (Join Operation: One to One, Match Option: Intersect). This created a new output feature class, hom_morans_sj,  in which I added a float data type field named “Crime_Rate” and Calculated Field with the query:

 “Crime Rate = !Join_Count! / !total_households! * 1000”

to determine the number of homicides per 1000 households.

Once the number of homicides per 1000 households was determined, I used the Cluster and Outlier Analysis (Anselin Local Moran’s I) Spatial Statistics tool using the hom_morans_sj and the input field being the created data field “Crime_Rate”.

I then used a SQL Query to select only the high-high (HH) clusters of crime, exported that selection as a new feature class, and used Dissolve.  

 

    Based off the results of each of the three hotspot mapping methods (Grid Overlay, Kernel Density, Local Moran’s I), I believe Kernel Density to be the best overall method in this situation to predict future homicides. Even though Kernel Density has less total square miles hotspot area in 2017 homicides, it still has a reasonable percentage of area overlap between 2018 and 2017 homicides. The less square miles of hotspot coupled with the reasonable percentage of crime overlap between 2017 and 2018 hotpot area makes Kernel Density the better choice of the option and would allow for the allocation of resources in a more realistic manner.