search

hackoregon / emergency-response

Simulations, Models, and Visualizations of Portland Fire and Rescue data
11 stars 10 forks source link

Compute false alarm metrics #25

Open futurechris opened 7 years ago

kielejocain commented 7 years ago

Blocked by #1

sky-t commented 7 years ago

I updated my notebook (https://github.com/sky-t/hack-or-emergency-response/blob/master/fire_dispatch_eda_incidents.ipynb) that looks at the incident description and incsitfoundclass for the full dataset. I'm happy to continue working on this issue in that notebook. What other aggregations regarding false alarm would you like? By year, by FMA, by fireblock?

hannahkane commented 7 years ago

Let's look at false alarms by FMAs.

@markawhitaker - we're trying to remember what your interest in false alarm data was. Can you help us recall if you had a specific question here?

markawhitaker commented 7 years ago

@hannahkane The general goal is to better understand the fire alarm "problem." Is there a rationale for creating policies to reduce commercial fire alarms, such as they occur so often that it's becoming a workload issue or affecting PF&R's ability to respond to other incidents?

Are false alarms concentrated in certain FMAs that would allow for targeted education and or enforcement in that area?

hannahkane commented 7 years ago

Thanks, @markawhitaker. This is a really useful problem statement. We'll see if we can tackle it.

rcallihan commented 7 years ago

I am working on a notebook for this issue. Can be found here.

qingdatascience commented 7 years ago

I did another analysis using tree base model to screen those categorical variables in the incident table. The preliminary results showed that those auto-alarms are the major culprits. I went ahead to use it to search the fire blocks, and listed top 10 relevant fire block codes. Hope this information combined with Ryan's analysis can shed some insight into this issue. Notebook is here

rcallihan commented 7 years ago

Nice work, @qingdatascience! It would be nice to distill our findings into a few key pieces of digestible and useful information to PF&R. For example, we should translate the results of the random forest classifier to language meaningful to the PF&R. I'll hopefully work on this Saturday. I'd be happy to use your help, if you're around.

qingdatascience commented 7 years ago

Sure. I can spare some time on Saturday.