Loan prediction

The goal of this project is to predict if a loan acquired by Fannie Mae will go into foreclosure or not.

Fannie Mae buys loans from lenders to try to incentive them to issue more loans.

Fannie Mae publishes here data for the loans that it has acquired and how they perform through time.

This project was done following a tutorial from Dataquest.

In 2020, Fannie Mae modified the way they publish their data. To avoid having to make too many modifications to the tutorial project, the old dataset was used.

An article on this project is available here.

Installation

Download the data

• Clone this repo.
• Go to the folder: cd loan-prediction.
• Create a data directory: mkdir data.
• Go to the data directory: cd data.
• Get the data:
  • Manually:
    • Download the 2000-2015 dataset here.
    • Extract the files from 2012 Q1 through 2015 Q1.
    • Remove the .tar file.
    • Move the files to the data directory.
  • Or run the download script: python download_data.py.
• Go back to the loan-prediction directory: cd ..

Install the requirements

• Install the requirements: pip install -r requirements.txt.

Usage

• Create a directory for the processed datasets: mkdir processed
• Combine the Acquisition and Performance datasets: python assemble.py:
  • This will create two files in the processed directory: Acquisition.txt and Performance.txt.
• Generate the training data: python annotate.py:
  • This will use Acquisition.txt and Performance.txt to create training data.
  • It will create a file train.csv with this data.
• Run cross-validation on the training set, and print the associated error metrics: python predict.py.