Algorithmic Bias and Fairness in Machine Learning Systems: A Review

Abstract

Machine learning (ML) systems are increasingly deployed in sensitive domains such as healthcare, finance, and criminal justice. Despite their benefits, these systems often exhibit algorithmic bias, raising concerns about fairness, accountability, and trust. Bias may emerge from historical inequalities in training data, model design, or deployment practices, resulting in disparate impacts on marginalized groups. Over the years, researchers have proposed multiple fairness definitions and mitigation strategies, ranging from data preprocessing and in-processing adversarial learning to post-processing adjustments. Toolkits like AI Fairness 360 and Fairlearn support practical implementation, while regulatory frameworks such as the EU AI Act emphasize governance and accountability. This survey consolidates theoretical foundations, technical approaches, domain-specific applications, and policy perspectives to provide a comprehensive understanding of algorithmic bias and fairness in ML, highlighting open challenges and future research directions.