Using Hotspot Analysis for Crime Prediction

For this exercise, data was provided for crimes in Washington D.C. and Chicago.  For the Washington D.C. analysis, we were walked through building a chloropleth and kernel density map to identify areas of chronic crime.  Each map began with lots of point features representing the locations of crimes.  For the chloropleth map, burglaries were filtered out using a SQL query and a spatial join was used with a feature class of census tracts to determine the number of burglaries within each tract.  A field was then added to the new feature class to determine the crime rate of each tract rather than only the number of homicides per tract (join_count / total # of housholds * 1000).  Once the outliers are determined and excluded, breaks are put in and the resulting map is very easy to understand which area are susceptible to burglaries. 

For the kernel density map, a raster is needed.  Before using the KERNEL DENSITY tool, Environments of the map need to be adjusted to the provided feature class representing the boundary of Washington D.C.  After determining the average value, breaks are manually put in by a multiplying factor of the average value.

For the grid-based thematic map of Chicago homicides, the provided feature class of ½ mile grid cells were used.  The first step in this analysis is to determine how many homicides unfortunately occurred in these areas.  By using a spatial join, ArcPro calculated how many point features representing the locations of homicides occurred in each cell.  Thankfully, there are cells without homicides and these cells were excluded from this analysis by selecting the cells with homicides by attribute i.e. any cell with a value join value greater than zero.  Of the 1,076 total cells, 311 cells had homicides occur within them.  Simple arithmetic (311 total cells containing homicides / 5 to represent the top 20% = 62.2) and sorting of the attribute table of the new feature class was used to determine the top 20% of grid cells with the most homicide points within (62 cells).  An analyst would want this information as one polygon, and to use the DISSOLVE tool, an attribute with an identical value is needed.  To accomplish this, an arbitrary field was created and subsequently used with the DISSOLVE tool to merge the individual cells together into a single polygon.

To create a kernel density map, a raster is created from the polygon created with the DISSOLVE tool by using the KERNEL DENSITY tool.  By viewing the statistics of the resulting symbology, two breaks are created based on the mean value (2.88 * 3 = 8.64) and the maximum value (38.92) in order to reclassify the raster back into a polygon using simple Boolean values.  By isolating the areas between 8.64 and 38.92, high crime areas are identified.

To identify high crime areas that are adjacent to other high crime areas, a Local Moran’s I analysis is used.  After performing a spatial join between the census tracts and total homicide point features, a new field is created to determine the crime rate of the census tracts.  Using a SQL query to divide the join count of the census tracts by the total number of households and then multiplying the result by 1000, the field calculator populates the crime rate field with the crime rate.  After running the CLUSTER AND OUTLIER ANALYSIS (ANSELIN LOCAL MORAN’S I) tool, high-high clusters are identified, isolated, and dissolved into a single polygon feature.

I believe the kernel density map would be the most efficient map for allocating police resources.  The total area of the kernel density map of Chicago homicides was larger than the grid overlay and smaller than the Local Moran’s I hotspot map though it had the highest density of homicides within a square mile than the other analyses.  Also, approximately 43% of the homicides from 2018 occurred in this area relative to approximately 45% of homicides occurring in the much larger Local Moran’s I hotspot map.  By focusing more attention to the areas within the kernel density hotspot map and working outwards, this most efficiently brings attention to the areas with chronic homicide.  Rather than allocating resources to the areas of the grid overlay and Local Moran’s I hotspot map immediately, working into those areas from the areas of the kernel density map may prove to be the best strategy in combating violence in Chicago.