waste_not_the_water

waste_not_the_water is data science tool for urban waste water treatment plants created through the DIRECT courses at University of Washington. This tool can be used to predict capacity for a new European wastewater treatment plant and a visualization of the database from the Waste Water Treatment Directive. waste_not_the_water utilizes the open Urban Waste Water Treatment database from the the European Environment Agency. Our primary use case is to create a data science tool that predicts capacity based on user inputs; this was done with the use of a user interface. waste_not_the_water also contains interactive data visualization showing the size and location of the existing waste water treatment infrastructure in Europe.

Description of the model

Regression Model

The features of our model are the load entering and the location in latitude and longitude. There exists a linear relationship between the capacity and the load entering the treatment plant, and we implemented simple linear regression. However, after examining multilinear statistics, it can be seen that latitude and longitude also affect the capacity. A Ridge Regression model was used due to the high variance in the data. The R² statistic for Ridge Regression is 0.91.

Note on units: capacity and load entering are given in population equivalent, a common water treatment unit that is normalized based on the amount of water usage for one person in 24 hours.

Nearest Neighbors

To give the user more information, a nearest neighbor algorithm was implemented that will return the two closest plants to the user's inputs. This is done through a simple distance calculation of weighted capacity and location coordinates. The phosphate and nitrogen removal are used as filters to further specify the treatment plants.

Data Visualization

The interactive maps were created with folium and mpl_toolkits.basemap. The figures allow us to clearly identify the area where most waste water treatment plants are located in Europe.

Relative size by dot	Size by color

User Interface

The user interface was created with the use of Dash by Plotly. Dash uses a python background that is converted into html. Our app is run locally on the web. Users can input their proposed plant feautures on the app, and it returns the capacity through the Ridge Regression model. The app is called through the command python app.py. The app runs locally on the web through this url. For more information on Dash, check out their awesome website.

Files setup

• The python modules are in waste_not_the_water directory.
• The useful data are organized into data directory.
• The unit test files are in the directory called tests.

Dependencies you need

This is what dependencies you need:

• basemap
• folium
• sklearn
• statsmodels
• scipy
• dash by plotly

Installation

This is how you install it:

• $conda install basemap / pip install basemap
• $conda install folium / pip install folium
• $conda install scilit-learn / pip install -U scikit-learn
• $conda install statsmodels / pip install --upgrade --no-deps statsmodels
• $conda install scipy / pip install scipy
• $conda install dash==0.21.0/ pip install dash==0.21.0
• $conda install dash-renderer==0.11.3 / pip install dash-renderer==0.11.3
• $conda install dash-html-components==0.9.0 / pip install dash-html-components==0.9.0
• $conda install dash-core-components==0.21.0 / pip install dash-core-components==0.21.0
• $conda install plotly --upgrade / pip install plotly --upgrade

Examples

For more information on how to use our package, go to the examples folder.

Acknowledgments

We would like to profusely thank our wonderful instructor, Prof. David Beck, and our amazing teaching assistants, Nick Montoni, Arushi Prakash, and Coco Mao.