Open meilinger opened 5 years ago
The next logical step then would be to tie stations that exist inside those boundaries to the departments.
For transparency, the next step with this is to cull out outlier incidents and perform a % inside vs outside of government units that are "close" to the department headquarters (or close to the median centroid of geocoded incidents associated w/ the department). Ideally, the govt unit w/ the highest percentage of contained incidents would be geospatial object chosen as the boundary for the department.
Prove this out using existing verified boundaries litmus test for the calculus.
In order to prove out the ability to use government units for department borders, we need to determine the fit of existing government units to known-good (eg. verified boundaries).
IPython notebook: For all departments that have a verified boundary, find all government units that intersect with the department headquarters (buffered by an amount). For each of those government units, sum the # of intersecting geocoded incidents and divide by area of the government unit. Government units with highest incident/area density will be ranked highest as candidates for department jurisdiction boundaries and should be compared against the existing validated jurisdictional boundary for our test set (% difference in area, % difference in centroid). Ideally, the most-dense candidate government unit will match the validated jurisdiction exactly if we hope to make this process programmatic.
If this test proves mostly-accurate, then we can continue to flesh out a system to more easily assign government units as boundaries by humans (and/or programmatically).
@cweinschenk @garnertb this was what I was thinking.