Accuracy and Fairness: An Analysis of Risk Assessment Algorithms in the Criminal Justice System

Abstract

Risk-assessment algorithms with black-box decision-making procedures are becoming increasingly prevalent in the criminal justice system in the United States. However, their advances into the justice system have not come without controversy. Specifically, there is significant concern about whether these algorithms are racially biased. The goal of this thesis is to investigate the fairness and accuracy trade-off of one of these risk assessment algorithms, the Correctional Offender Management Profiling for Alternative Sanction (COMPAS). Specifically, using different data transformations suggested by the literature, it will try to un-bias biased historical data and achieve a better trade-off between fairness and accuracy for black and white defendants. Finally, it will propose two transformations that achieve better fairness without compromising accuracy and discuss the scope and limitations of these transformations