Comparing Methods For Detecting And Mitigating Bias In Machine Learning Models
Akash Bhise, Snehal Mali, Prerna Joshi, Saurabh Mohabe, Diptee Vishwanath Chikmurge, Sunita Barve · 2023
As these models have the potential to both reinforce and generate new inequities, bias in machine learning models has a substantial influence. It is important to address and mitigate bias for having fairness in the outcomes. This study intends to investigate the area of bias in machine learning models and investigate several methods for its identification and mitigation. An adult income dataset is used, which contains variables like age, education, occupation, and gender that may have an impact on a person's income. Evaluation is done using various techniques for detecting bias, including statistical measures such as demographic parity and equalization of odds. Our research shows that bias, particularly with regard to gender and ethnicity, is a serious issue in machine learning models trained on the adult income dataset. This study also highlights the need of carefully selecting appropriate measures for evaluating how fair machine learning algorithms are. This study emphasises the necessity for machine learning professionals to carefully address the issue of bias in their models and pick acceptable methods to identify and reduce it. In the results it was observed that statistical parity techniques are efficient for mitigating bias. In this way, it shows that machine learning models are fair and can be used to promote social and economic justice in a variety of fields.