Math models and policing

[lwt16]

Modeling also shows times.

I got my list of "hot spots" for violent crimes that officers are supposed to mark out at and be visible for 15 minutes a couple times a day. One is somewhere I *know* there is no violent crime. It's an area with a bunch of camp grounds, hiking trails, horseback riding places, etc. It's also the home to a bunch of unofficial, but not illegal, ranges. Neighbors call in "shots fired" runs all the time because of guys using those ranges.

As a sergeant, the first thing I ask after reading these things at roll call is to the guys that work that area if it makes sense and what additional info they have just because I know the automated data results in that sort of thing on occasion.

This is an issue we struggle with anytime a hunting season kicks off out in our county (multiple counties now) JD. We have several high end neighborhoods that were built out there and due to annexation, are now part of our metro area. Dove, squirrel, deer, etc season kicks off and all those people in those neighborhoods absolutely freak out and light up 911 with "shots fired" calls.

If you looked at it on paper, you'd think it was a war zone. The zone units roll their eyes and start with the "Karen" jokes. All of them have plenty of stories on how they had to explain to people that yes, it is legal to hunt game in the fields that back up to their yard, and no, we aren't going to go "have a talk" with people engaged in legal activities just because you didn't recon where you signed a 500,000 k mortgage.

It's always best to ask the zone units. They know where the hot spots are and who the players are.

Regards.

[LtDave]

Modeling is wonderful, but if you can barely staff your patrol beats, where are these “extra” patrols of hot spots coming from?

[jh9]

Modeling is also stupid, in the "garbage in, garbage out" meaning. I got my list of "hot spots" for violent crimes that officers are supposed to mark out at and be visible for 15 minutes a couple times a day. One is somewhere I *know* there is no violent crime. It's an area with a bunch of camp grounds, hiking trails, horseback riding places, etc. It's also the home to a bunch of unofficial, but not illegal, ranges. Neighbors call in "shots fired" runs all the time because of guys using those ranges.

Do you not have the ability to even indirectly change the class label for a false positive? Eventually the model should pick up on that. If your feedback (what's right, what's wrong) isn't included with new input data the (supervised) modeling component will just be throwing darts.

perlslacker is right in that there's a lot of garbage in that space. I've written some of it. On some projects the people we sent out to train the users didn't know how the software worked themselves. So they just helped the one or two people in the customer's org that had papal dispensation to touch the software learn how to spin the knobs and turn the dials until they got a something that looked kind of right. The results would end up as a word/excel/powerpoint that the actual people using the results don't find useful. Communication between people who used the predictive data and the dev team always went through 3 or 4 intermediaries, none of whom understood much about either the problem domain or the math behind it all so it ended up being a game of telephone.

Wasted time and wasted money.

[BehindBlueI's]

Do you not have the ability to even indirectly change the class label for a false positive? Eventually the model should pick up on that. If your feedback (what's right, what's wrong) isn't included with new input data the (supervised) modeling component will just be throwing darts.

Short answer: Probably not.

[jh9]

Short answer: Probably not.

I can't imagine that ever generating useful results.

[Doug]

It started with GIS (geographic information system) basically a map with a database. (Think Digital pin map)

From my experience, then the nerdy type police officers (sorry) and the developers sought out the good idea mantra....

Much and many dollars later, complex systems Came about that supposedly help. Chiefs were sold on their value and their dashboards.

Yet to see a good study that compares different size agencies (small, medium and large) that evaluates benefits/cost savings between those using the systems and those not.

Waiting for real time drone data acquisition, camera integration, license plate readers...If it has not progressed that far already.

Future AI (artificial intelligence) will be a rude awakening for the politicians when OCP (90’s reference) tells them they don’t know why the AI makes the decisions it does.

[KellyinAvon]

It started with GIS (geographic information system) basically a map with a database. (Think Digital pin map)

From my experience, then the nerdy type police officers (sorry) and the developers sought out the good idea mantra....

Much and many dollars later, complex systems Came about that supposedly help. Chiefs were sold on their value and their dashboards.

Yet to see a good study that compares different size agencies (small, medium and large) that evaluates benefits/cost savings between those using the systems and those not.

Waiting for real time drone data acquisition, camera integration, license plate readers...If it has not progressed that far already.

Future AI (artificial intelligence) will be a rude awakening for the politicians when OCP (90’s reference) tells them they don’t know why the AI makes the decisions it does.

I think OCP dates back to the 80s. AI will do what it's trained to do. Until the machines try to kill us like in the movies...

[Doug]

I think OCP dates back to the 80s. AI will do what it's trained to do. Until the machines try to kill us like in the movies...

Was referring to the new AI that has bested the strategy game GO. It is not like there is a line of code since it teaches itself to play. Then learns as it plays over and over.

I think we are a long away off before the killing machines. We will crash the markets long before then.

Thank, you are right, RoboCop was the 80’s