A Survey on Bias and Unfairness in Machine Learning Models

Bias and unfairness in machine learning models are serious issues that can affect the quality, reliability, and ethics of artificial intelligence (AI) systems and applications. In this survey, we provide a comprehensive overview of the current state of the art on how to detect and mitigate bias and unfairness in machine learning models. We follow the PRISMA guidelines for conducting a systematic review and search for relevant articles in four major databases: Scopus, IEEE Xplore, Web of Science, and Google Scholar. We select ۴۵ articles that meet the inclusion and exclusion criteria and analyze them based on four aspects: datasets, tools, fairness metrics, and identification and mitigation methods. We identify the challenges and limitations of the existing datasets, tools, metrics, and methods, as well as the trade-offs and conflicts among them. We also discuss the different sources and types of bias that can affect AI applications in various domains and subdomains, such as computer vision, natural language processing, recommender systems, and others. We summarize the different fairness definitions and criteria that machine learning researchers have proposed to avoid or reduce bias and unfairness in machine learning models. We highlight the future directions and challenges for research on bias and unfairness in machine learning models. We hope that this survey will serve as a valuable resource for researchers, practitioners, policymakers, and stakeholders who are interested in understanding and addressing bias and unfairness in machine learning models.

Keywords:

Keywords: bias , unfairness , machine learning , artificial intelligence