Alberta Oil Spills Analysis

Contributors

• Alycia Butterworth: alyciakb
• Huijue (Juno) Chen: huijuechen

Latest

• Date: December 8, 2018
• Release v4.0

Overview

The goal of this project is to find the top predictors of the cause of a spill incident in the Alberta oil industry. Using data from the Energy Resources Conservation Board (ERCB), provided by the City of Edmonton, we reviewed spills that occurred between 2002 and 2013 and fit a machine learning model (decision tree) to predict spill cause as either (1) Equipment Failure or (2) Operator Error based on five factors: (1) the field house location of the spill, (2) the time of year it occurred, (3) the source of the spill (well, pipeline or battery), (4) the substance spilled (oil, gas production, or water) (5) and the volume spilled. We were able to obtain a gini score for each factor and rate them accordingly.

We will use the following contents from the data:

Our findings showed that the top three predictors for the cause of an oil spill in Alberta are:

1. source of spill
2. field location
3. type of substance spilled

The full final report that includes our analysis and interpretation, limitations, and future directions can be found within the doc folder. The report also summarizes and visualizes our original data, visualizes the decision tree model created for this analysis, discuesses its accuracy and the results. This analysis can be fully reproduced by a user within their terminal. Please see the Procedure section below for instructions.

Procedure

To run this analysis yourself, the scripts need to be run in the following order:

1. Cleaning data in 1_data_cleaning.R
2. Visualization of data in 2_data_viz.R
3. Use sklearn to fit a model in 3_model_fitting.py
4. Visualize the model decision tree in 4_model_viz.py
5. Create a final report and render it in final_report.Rmd

The output files creating when running this report are:

1. A clean_data.csv file that is used for the rest of the analysis.
2. Six graphs, saved as png files, stored in the img folder that help the user visualize the data we are using for the analysis.
3. Six small dataframes stored as csv files in the results folder.
4. An image called depth_compare.png, that shows the results of the 10-fold cross validation, saved in the results folder.
5. The decision tree code model, saved in the results folder as final_model.sav.
6. A visual representation of the decision tree, called spills_tree_model.png a png file in the results folder.
7. Rendered markdown and rendered html versions of the final analysis report, saved in the doc folder.

Reproduce with Docker

This report can be reproduced using Docker. To run the analysis using Docker:

1. Clone/download this repository.
2. Navigate to the root of this project in your terminal.
3. Type the following command in your terminal (filling in PATH_ON_YOUR_COMPUTER with the absolute path to the root of this project on your computer):

docker run --rm -v PATH_ON_YOUR_COMPUTER:/home/alberta_oil_spills alyciakb/dsci_522_alberta-oil-spills make -C '/home/alberta_oil_spills' all

4. To clean and remove files created by running the analysis:

docker run --rm -v PATH_ON_YOUR_COMPUTER:/home/alberta_oil_spills alyciakb/dsci_522_alberta-oil-spills make -C '/home/alberta_oil_spills' clean

Reproduce with Makefile:

This report can be run in the Docker environment or without Docker. To run this report directly from your terminal, we have provided a Makefile that will run the scripts in the correct order and produce all output. To run the analysis using the Makefile:

1. Clone this repo.
2. Navigate to the root of this project in your terminal.
3. Type the following command in your terminal: make all.
4. To remove files created by the Makefile, type the following command in your terminal: make clean.

Dependency diagram of Makefile:

Dependencies

R & R Libraries:

• R (version 3.5.1)
• rmarkdown (version 1.10)
• tidyverse (version 1.2.1)
• lubridate (version 1.7.4)
• gridExtra (version 2.3)

Python & Python Libraries:

• Python (version 3.7.1)
• numpy (version 1.15.1)
• pandas (version 0.23.2)
• seaborn (version 0.9.0)
• matplotlib (version 3.0.2)
• scikit-learn/sklearn (version 0.20.1)
• graphviz (version 0.10.1)

Release History