Predicting the Risk of Recidivism among people released from prison in North Carolina

Spring 2019

Vedika Ahuja, Pete Rodrigue, Bhargavi Ganesh

Paper

Our final paper summarizing the results of our models and outlining the objective of the project is here.

Data

• (North Carolina Deapartment of Public Safety Offender Public Information database)[https://webapps.doc.state.nc.us/opi/downloads.do?method=view]

Contents of Repository

• aequitas_analysis.py: This script runs the aequitas analysis on bias in our pipeline
• best_model.py: This script takes the best model (as determined in run_pipeline.py) and creates a feature importances, precision/recall curve, and a decision tree stump
• data_explore_library.py: This script contains functions for data exploration
• data_explore_script.py: This script creates plots/tables in data exploration
• ml_functions_library.py: This script contains the functions used in the ml pipeline, to create evaluation tables
• processing_library.py: This script processes the data and creates a dataframe to be fed into the ml pipeline
• run_pipeline.py: This script runs the ml pipeline, using a set of parameters that are looped over for various different models