COMPAS Recidivism Score Predictor

Inspiration

We aim to disrupt the criminal justice system by introducing more transparency to the AI predictive tools used in court cases.

What it does

We suggest a system that approximates the scores that a certain Pre-Trial Risk Assessment Instrument (PRAI) in a defendant's region would give them based on their information, and allow them to use that information for a trial. Their lawyer can also find and use that information. More on this can be read on our whitepaper on the deployment page. Here is the link to the whitepaper: https://docs.google.com/document/d/1Ur949GM5mvjZjpTAjn-Cnyd4nKmzX1He6tdusI5oNS0/edit?usp=sharing)

How we built it

We built a Machine Learning model that aims to replicate the decisions of the COMPAS model (since we cannot have access to it). It is a GradientBoosterClassifier from the sklearn library.

Challenges we ran into

Tweaking the model and finding the right parameters, and right features was a long process.

Accomplishments that we're proud of

The model currently mimics the COMPAS predictions at about 67% of the time. This is above random chance and we are confident can be improved.

What we learned

There is so much lack of transparency and shaky science in the prediction of recidivism, and it's surprising that it's not talked about more given how much it impacts the lives of defendants, especially those of colour.

What's next for COMPAS Recidivism Score Predictor

We hope to improve the model's accuracy, and do more independent research on the matter. Hopefully, this can eventually translate into a paper that can help others consume this knowledge.

Built With

• pandas
• sklearn
• streamlit