Recidivism Prediction

This repository documents the project Aaron Dunmore and I completed in the fall of 2019. We developed risk assessment instruments (RAIs) predicting two-year recidivism in Broward County, FL, using the data released by ProPublica. We evaluated the predictive performance of our models and compared it to the performance of the RAI used by the courts in Broward County previously. In addition, we investigated whether our RAIs are equally predictive across race/ethnicity, age, and sex groups.

This project was completed as part of Data Mining taught by Prof. Alexandra Chouldechova.

Instructions

This repository contains four folders:

Folders	Description
Notebook	Final report. You can execute it without having to run any other code
Data	Processed compas.db and .csv files used by our models
Cache	Trained models as R objects. It will save time when you re-excute our notebook
Data_Processing	.R and .sql files used to generate processed compas.db and .csv files in Data

You can also rebuild our project from scratch:

1. Delete the contents of Data. Place compas.db in Data.
2. Open compas.db in a sqlite manager, and execute Data_Processing/data_processing.sql. We used DB Browser for SQLite. This will create a number of variables used by our models in the database.
3. Execute Data_Processing/data_export.R. Your working directory should be set to the top level of this folder. This will form several .csv files used by our analysis.
4. Execute Notebook/final_report.Rmd. Your working directory should be set to Notebook.

If you would like to train all models as new, delete the contents of Cache.