Files in this item



application/pdfhao_poster.pdf (6MB)
(no description provided)PDF


application/vnd.openxmlformats-officedocument.wordprocessingml.documenthao_abstract.docx (142kB)
(no description provided)Microsoft Word 2007


Title:Fairness of machine learning applications in criminal justice: Insights from evaluation of COMPAS
Author(s):Hao, Yue
Subject(s):Machine Learning
Abstract:Machine learning has been widely applied in facilitating high-staked decision making, however, there is an increasing concern on hidden biases behind these methodologies. In the criminal justice context, there is a lasting debate on the fairness of Correctional Defendant Management Profiling for Alternative Sanctions (COMPAS) which uses Random Forest as foundation for recidivism risk predictions. But we noticed that fairness of the algorithm is genuinely measured by two different criterion: calibration and equalized odds. In this research we trained a Random Forest classifier and examined why its application is not eligible in achieving fairness based on different measures. Results show that both of the scales were not achieved on all the six racial groups in COMPAS data set which calls for further evaluation on the algorithm design and more efforts in defining a universal definition and measuring standard on algorithms fairness.
Issue Date:2022-02-28
Genre:Conference Poster
Rights Information:Copyright 2022 is held by Yue Hao. Copyright permissions, when appropriate, must be obtained directly from the author.
Date Available in IDEALS:2022-04-22

This item appears in the following Collection(s)

Item Statistics