Applications Module 1

 

GIS Applications

Module 1

Crime Hotspot Analysis

In module 1 the lab the goal was to perform multiple hot spot analysis processes within ArcGIS Pro using multiple methods that are commonly used to predict crimes within certain areas based on prior locations of the crimes committed. 

    The data provided was a mix of polygon and point file classes and was separated between two locations. One group of file classes were located in Washington DC and the other group of file classes were located in Chicago Illinois. Both sets of shape files were describing data gathered in 2017 and 2018. 

 

 

The first lab process was looking at crime rate data and census tract data in Washington DC to perform an analysis of the highest occurrence of burglaries per 1000 homes. This process results in a map with aggregated data instead of looking at raw data because it takes into account the amount of residential structures relative to burglary rates. The geoprocessing tool used for this process was a spatial join.
The result of this spatial join analysis is below:

The next Crime hotspot analysis considered a different criminal offense. Instead of burglaries we took a look at offenses labeled as "Assaults with Dangerous Weapons". Along with using the geo-process of spatial join, we also converted the vector layer created from the join into a raster layer using the Kernel Density Analysis which was produced from the assaults feature class created from the spatial join. 

The symbology was changed to represent varying levels of crime rates that excluded zero values. 

Below is the resulting map of the Kernel Density Analysis: 

 

 

 

 The next part of the lab focuses on homicides in Chicago, Illinois. In the next three steps we used three different analysis methods which include a Grid Overlay, Kernel Density Analysis, as well as a Local Moran's I Analysis. 

Many of the steps starting out begin with the same processes. Some of those steps include table joins, spatial joins, and new fields added to those tables in order to create a crime rate which incorporates an SQL query within the calculate fields function. 

The first analysis of implementing a Grid overlay required us to take the total homicides in 2017 point file for Chicago and join it to the Chicago Grid polygon file using the "CONTAINS" method. 

This creates a new polygon file with a join field that shows how many points of data were contained within each grid. Once that was created we then took that new polygon field and manually selected the top 20% of grids with the highest crime rates and exported that too a new table. 

Once this was accomplished we could then overlay the Total Homicides for 2018 point file to see how accurate this simpler method could predict future crimes of homicide. 

Below is a screenshot of what that looks like. 

The total Area of the 2017 homicide grids equals 14.46 (mi*2) and matching that up with the 2018 homicides data points, only 159 out of 589 homicides in 2018 could be considered predicted in relation to the prior year of 2017. I would not consider this an highly accurate or even moderately accurate analysis for predicting most likely crime rates in this specific instance because the grid over lay only captured 26.9% of the crimes that would occur the following year. 

    However, this method of analysis did out perform the following two that I will discuss when it came to the (Crime Density within square mile grid hotspot) at a rate of 10.28. This seems to be a mater of chance though because a large portion of the homicides in 2018 occurred in a tightly packed area in the north western section of Chicago. 

As a final caveat, this method may be more useful than the other two analysis methods if the intended goal was to determine the areas with the highest density of homicides within the city out of all homicides.  

The second analysis method implemented the kernel density analysis tool within ArcGIS Pro. 

To do this, we repeated the same steps from the previous kernel density map that was discussed above this section but used the data for Chicago Illinois instead. 

    After the Raster output was created we then set the breaks for the symbology to (two breaks) instead of six like in the previous map. This difference drastically changes the hotspot area appearance in that there isn't a cold, warm, hot, very hot, sort of feel to the hotpot area; rather a general outline of crime in one color. This was reclassified and eventually processed to turn the raster layer back into a vector layer. As raster layers do not contain a attribute in ArcGIS we needed to know the total area of the hotspots from the prior steps. 

Once these steps were completed, we then added in the 2018 homicide point file to then run through a couple of more processes to learn how accurate this analysis method turned out to be.  

Below is a screenshot of the map that was used to analyze accuracy of predicting future homicides in Chicago. 

The results learned from this analysis method were as follows:

Total Area of 2017 hotspot: 47.80 (mi*2) 

Number of homicides within hotspot area: 378 out of 589 

% of 2018 homicides within hotspot area: 64.2%

Crime density within 2017 hotspot area: 7.91

The final analysis method being very similar to the previous two methods in data processes performed used an analysis method called Local Moran's I. This method aggregates the data being examined and in this last step we used a census tract polygon file for Chicago. 

For this aggregation we divided the amount of homicides within the census tracts per 1000 housing units.  When the (Anselin Local Moran's I) spatial statistics tool is ran, it creates different clusters of data which in this case would be homicides in the year 2017. 

Those clusters are broken in to LowLow, LowHigh, HighLow, HighHigh. 

We of course only wanted to view the HH cluster so we exported those tracts considered HH to a new feature class and then overlayed the 2018 Homicides.

Below is the output:

 

The results are as follows:

Total Area of 2017 hotspot: 35.12 (mi*2)

Number of homicides within hotspot area: 271 out of 589

% of 2018 homicides within hotspot area: 64.2%

Crime density within 2017 hotspot area: 7.74

My final conclusion for the analysis methods performed are as follows:

As the hotspot total area increase, so does accuracy. This is an obvious statement as if the entire city of Chicago was a hotspot layer the accuracy would be 100%

Further calculations would need to be performed to compare the accuracy between all three and percent of 2018 recorded homicides contained within the 2017 hotspot areas.