Switch branches/tags
Nothing to show
Find file History
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
..
Failed to load latest commit information.
notebooks
reference
sources
tools
2016-spr-mjclawar_rarshad-poster.pdf
2016-spr-mjclawar_rarshad-report.pdf
README.md
__init__.py
crime_knn_weekday.html
home_value_model.html
hospital_scatter.json
plot_hospital_scatter.html
requirements.txt
run.sh
school_scatter.json
setup.py

README.md

mjclawar_rarshad

Dependencies

Running the package

  • Create a file named auth.json with fields
{
    "services": {
        "cityofbostondataportal": {
            "service": "https://data.cityofboston.gov/",
            "username": username,
            "password": password,
            "token": token (API token for the City of Boston portal)
        }
    }
}           
  • Run setup.py auth.json

Reference information

The Provenance document for our project can be found in reference/plan.json A depiction of the network graph is provided in the svg reference/provenance.svg

Data Sets

  1. Crime Incident Reports link: http://https://data.cityofboston.gov/Public-Safety/Crime-Incident-Reports/7cdf-6fgx

  2. Property Assessment 2015 link: https://data.cityofboston.gov/Permitting/Property-Assessment-2015/yv8c-t43q

  3. Hospital Locations link: https://data.cityofboston.gov/Public-Health/Hospital-Locations/46f7-2snz

  4. Boston Public Schools (School Year 2012-2013) link: https://data.cityofboston.gov/dataset/Boston-Public-Schools-School-Year-2012-2013-/e29s-ympv

We are interested in evaluating Boston neighborhoods/properties based on a variety of parameters. Important parameters include incidents of crime and proximity to facilities such as schools and public transportation. We will ideally include additional relevant data sets such as emergency response times, demographic data, and more. We will come up with a value metric based on our data. We will then compare our value metric to actual property assessment data and observe the following:

- What parameters strongly influence property value?
- How well does our model agree with actual property data?

In a previous version of this project, we began building the foundation for neighborhood evaluation. In particular, we found "crime centroids", about which clusters of crime tend to have occurred in 2015: processing/crime_centroids.py

  • The "optimal" number of crime centers is estimated using a naive rule of a 1% change in the Bayesian Information Criterion (BIC)

We also estimated the distance between properties in Boston and the nearest hospital using a spherical approximation: processing/hospital_distances.py

In this iteration, we used k-nearest neighbors regression with latitude and longitude as the design matrix for predicting smoothed housing values per square foot across Boston.

We also extend the crime centroids analysis by again using k-nearest neighbors regression on latitude and longitude to predict whether crimes at a given location are more likely to occur on weekends (Fridays to Sunday) or on weekdays (Monday to Thursday).

We plot both the housing values and crime classifications using Leaflet and OpenStreetMap and CartoDB map layers with an extended version of the excellent mplleaflet package. To view the home values data, run the package using setup.py and open home_value_model.html in a browser. Similarly for the crime predictions, open crime_knn_weekday.html in a web browser. Previously estimated versions of these two html files are in the reference folder and can be viewed in a browser.