aabrahamson3/tanzania-water-wells

# Predicting the Functionality of Water Wells This README.md file will serve as a roadmap to this repository. The repository is open and available to the public. ## Directories and files to be aware of: 1. An “environment.yml” file that contains the packages necessary to run the executables 2. An src/ directory that contains a .py module - In the root directory of this folder on your local machine, in your terminal please run 'pip install -e .' to allow the notebooks to access our functions. This will run our 'setup.py' file in the root directory 2. A notebooks/ directory that contains three Jupyter notebooks - A data exploration notebook - A modeling notebook, containing five models - A presentation notebook, containing our final report and model 3. A data/ directory containing three data files - Due to GitHub upload restrictions, these are included as .gitignore files. They are, in brief: - Training set target labels - Training set features - Test set features 4. A one-page memo.md written summarizing the models’ results, written for non-technical stakeholders 5. An “Executive Summary” slideshow PDF available as “Presentation.pdf” The data files described above can be found on drivendata.org. An account setup is required for download. A data dictionary can be found here: drivendata.org ## Methodology We performed a thorough EDA of the dataset, and built several models to detect if a water well is in need of repair. We tried 5 different classification models, with a Random Forest Classifier as the best performer. This had a higher overall F1 Score, as well as the best recall for the 'Needs Repair' category. Higher recall means fewer False Negatives - we believe that this is the best metric by which to evaluate the performanceof the model; a well needing repairs yet being labeled as “Functional” could have …