Lab 1: Crime Analysis

For our first lab of Applications in GIS, we learned all about hotspot mapping and most commonly used clustering methods. Below are three screenshots of three different types of crime hotspot mapping techniques for homicides in Chicago: 

Local Moran's I Hotspot Mapping

Local Moran's I hotspot mapping technique is based on whether features nearby are similar or not. To start the analysis, I completed a spatial join between the census tracts feature class and the total homicides 2017 data to get homicide counts by census tract. I added a new field to the new layer that was created from the spatial join and calculated the homicide rate based on the number of homicides per 1,000 housing units. I then used the Cluster and Outlier Analysis (Anselin Local Moran's I) tool with the homicide rate as the input field to get spatial autocorrelation patterns. The output classified each census tract based on the cluster type: high-high, low-low. I then used a SQL query to select only the high-high clusters to get areas with high homicide rates surrounded by other high rate tracts. Finally, I dissolved the selected polygons into a single multipart polygon using the COType field which is shown in the screenshot above. 

Grid-based Thematic Hotspot Mapping 

To create this hotspot map, I ran a spatial join to combine the total homicides point data layer with a 1/2 mile grid clipped to the Chicago city boundary. The spatial join created a new feature class with the count of homicides per grid cell. I then used the selection tool to only grid cells with at least one homicide and created a new feature class from this selection. I used a quintile classification approach by manually selecting the top 20% grid cells with the highest homicide counts using the Join_Count field that was created in the attribute table and creating a new feature class with the top 20%. I created a Dissolve field and populated each cell with the same value. Using he Dissolve tool, I merged the top 20% grid cells into a single multipart polygon. This newest polygon shows the most intense homicide hotspots in 2017.

Kernel Density Hotspot Mapping

I also performed a kernel density hotspot map using the total homicides in 2017 dataset. I ran the Kernel Density tool with no population field, a 100-foot output cell size, a 2630-foot search radius, and square miles as the area unit. I selected the planar method and I used the Chicago Boundary layer as a barrier to limit the calculations to the city boundary. This generated a raster layer representing spatial intensity of homicides across Chicago. I edited the symbology to display only values equal to or greater than three times the mean density value. Then I reclassified the raster into two value ranges - below and above the 3x mean quantity. I then used the Raster to Polygon tool on this new raster layer. I used Select by Attributes to export only the polygons with a grid value of 2 which showed they were above the 3x mean quantity. This is the final output polygon layer. 

I believe Local Moran’s I would be the most effect hotspot technique for predicting future homicides because it covers the largest area and it captures 267 homicides from 2018 which is 45.33% of all incidents. This is the highest amount from all three hotspot methods. Meanwhile, the Grid Overlay hotspot method captures 159 homicides in a smaller area – 15.46 sq mi. The Kernel Density has the highest crime density however, it only captures 98 homicides in an even smaller 7.24 sq mi area. For a police chief managing allocating limited policing resources within a city, Local Moran’s I would be most advantageous for them to use. Even through its density is lower, it can identify almost half of the city’s homicides which will help with future hotspot predictive and proactive policing.