Results for a “Hunch” set by a user with crime points overlaid on a heat map that shows crime density in a geographic area of interest. The Time-of-Day/Day-of-Week chart below shows crime density at various points in time at that location.
The purpose of HunchLab, Azavea’s geographic early warning system, is not only to identify changes in the geographic patterns in crime and alert users about detected crime spikes, but also to help users analyze and make sense of information about those spikes.
The “Hunch Detail” section of the application provides several tools geared for exactly that purpose. For the past several months we have made significant changes to it. In addition to thematic mapping and Time-of-Day/Day-of-Week graphs, the latest upgrade provides a heat map layer. This feature provides a visualization of crime density in a hunch’s geographic area during a specific timeframe.
When designing this new feature, we faced the question of how to best determine density. One way would be to divide an area into square cells and take a simple count of crimes per cell, but these results may not give a clear indication of actual crime density. If there are two different cells, neither of which contains any crime points, then the density values for both will be zero. But if one is in the middle of an area with no crime, while the other is adjacent to a cluster of crimes, this difference should really be reflected. So HunchLab makes use of a more sophisticated process known as “kernel density estimation” which overcomes this problem
“What is kernel density analysis?” you may ask. Well, rather than treating each crime event as if it only impacted the exact point where it occurred, the effect of each event is spread over the surrounding area. The distance over which the effect is spread can vary and is specified by a parameter known as the “search radius”. A “kernel” defines the pattern to use for distributing the event’s impact. Several kernels are available, each with a different pattern. The simplest kernel distributes the impact uniformly over the surrounding area, but most use mathematical formulas to give a greater weight closer to the crime event itself.
We’ve not only added this new density calculation feature to HunchLab but also made it into a reusable library that we can add to other applications. The new library leverages the Azavea Raster Grid (ARG) format already developed for Azavea’s DecisionTree product as well as other investments from previous projects. In HunchLab, users can adjust several parameters, such as the search radius to use during analysis and the rendering approach to use for displaying the results (equal interval or quantile). The heat map layer is just one of several tools available to help users analyze the crime spikes that HunchLab detects, with even more on the way!
