Measuring Bias

The extensive use of machine learning (ML) for supporting or making major decisions such as employment, credit card approval, or juridical decisions has resulted in rising concerns over the widespread existence of bias and discrimination. Decisions made by biased ML models result in damaging consequences for many people. In this paper, we focus on the effectiveness of statistical measurement methods to quantity the inherent data bias by examining several real-world datasets. The effects of bias on three classical ML techniques (Random Forest, Logistic Regression, and Support Vector Machine) are also studied to understand if bias affects their prediction results. To better understand the impact of bias on sensitive attributes, we also study the effects of bias on different protected groups with various correlation levels to the class labels.