proj-food_in_the_hood

Contents

• App Folder

  • manage.py - script for creating dashboard visualization with Dash
  • decision_tree.py - script for analyzing data using a decision tree classification algorithm
  • install.sh - shell script for creating the virtual environment
  • requirements.txt - text file of required libraries to install in virtual environment
  • regression.ipynb - jupyter notebook file for regression analysis
  • visualization_report.ipynb - jupyter notebook file for cumulative visualization report

• Data Folder

  • get_data_from_atlas.py - script for getting data from Chicago Health Atlas

  • food_data.csv - main file data file for dashboard visualizations and decision tree analysis

  • poverty_and_crime.csv - data file used for additional visualizations in Visualization_report jupyter notebook

  • several other draft csv files and draft jupyter notebooks for data visualizations

  • crime Folder

    • get_data_from_portal.py - script for getting data from Chicago Data Portal
    • total_crime.csv - used in visualizations in visualization_report.ipynb
    • several csv files related to crime data

  • project_visualizations folder

    • visualization_report.ipynb - cumulative visualization report
    • several other csv files necessary for creating visualizations

Running The Software

Language requirements: Python 3.8.5 Required Libraries: see requirements.txt

Code to run from the command line from within the app directory:

1. bash install.sh

2. source env/bin/activate

3. python3 decision_tree.py & python3 manage.py

4. Navigate to the address from the previous step, e.g. http://127.0.0.1:8500/, in a web browser.

5. jupyter notebook regression.ipynb and follow one of the links starting with http://localhost: (e.g. http://localhost:8888/?token=59fa90841a008fbc90400d4ebdca537ae241d9ced4f6f0cf) or http://127.0.0.1: (e.g. http://127.0.0.1:8888/?token=59fa90841a008fbc90400d4ebdca537ae241d9ced4f6f0cf) in a web browser and select "regression.ipynb"

   (Note: statsmodels package may need to be installed locally; for some group members this package was causing problems in loading the notebook)

6. (Optional): jupyter notebook visualization_report.ipynb and follow one of the links starting with http://localhost: or http://127.0.0.1: in a web browser and select "visualization_report.ipynb"

   (Note: You may need to run the notebook file chunk by chunk in order to see the visualizations because of how Plotly works)

Examples of Interacting with the Software

• The result from running decision_tree.py is a dictionary where the keys are the following strings: all variables, adult_fruit_and_vegetable_servings_rate, adult_soda_consumption_rate, low_food_access, poverty_rate, and population). The value associated with all variables is the performance rate for a model that uses all attribute variables in successfully predicting the crime rate. That is, how well did a model using all attributes in food_data.csv do at predicting the target class? We found that a model with all variables predicted the crime rate correctly about 55% of the time. Values associated with the remaining keys show how well a model did at predicting the crime rate correctly excluding that variable (key). For example, the poverty_rate key is associated with a value of about 41%, meaning a model built without the poverty rate attribute predicted the crime rate correctly 41% of the time.

• The result from running manage.py creates an interactive dashboard. You can interact with the dashboard to choose a variable and explore it in greater details by selecting from the left sidebar drop down menu. You will see a more complete description of the selected variable below the dropdown menu. For example,

• In addition to interactive visualizations, the dashboard also includes a correlation matrix heatmap, which visualizes the correlations between the following variables: adult fruit and vegetables servings rate, adult soda consumption rate, low food access, poverty rate, crime rate, and population

• The result from running the regression analysis substantiates our findings. Running a regression of crime_rate on adult_fruit_and_vegetable_servings_rate, adult_soda_consumption_rate, low_food_access, poverty_rate, and population shows that, again, poverty_rate is the most statistically significant explanatory variables and adult_soda_consumption_rate, though not significant, appeared to have some influence. Notably, a model with those only two variables succeeded in predicting crime rates with the highest R2, adjusted R2, and the lowest AIC and BIC.