Predicting Crime in Portland – data from the trenches

medium.com

18 points·magda_wang··8 comments

just skimming through, it looks like she's using past data to try to predict the future (call time and location, etc)

this seemed flawed from the start

it is an interesting project though

The aim of the challenge is to predict crime hotspots which for the most part don't vary much I'd think? The challenge then becomes identifying cluster density as a function of time which might mean using past data makes sense.

Why is it flawed. Crime hotspots are usually cyclical varying with seasons and weather like she points towards the end. There are other factors. You might be thinking of stock markets which is a different thing.

Doesn't that assume a stable population and a consistent economy?

I couldn't understand the argument, can you elaborate with some technical/scientific references ?

P.S: I am not the OP

> POIs are tagged with user-generated category tags such as “Amenity,” “Hotel,” “Shop,” “Restaurant,” “Mailbox,” etc. We found 11,000 labeled points of interest in Portland in the 2016 dataset.

Did you use the API or the shape file in here ?

Can you share the list of relevant Census feature to get the names, re-writing them from the picture takes time.

I hope they are thinking of the shortcomings of doing that: https://motherboard.vice.com/en_us/article/mg7w4x/canada-hub...

Good idea on using weather and holidays. I played with a similar dataset from local police for a machine learning class. And I remember how hotspots movement was correlated with seasons.

Reply on news.ycombinator.com