Bias and Fairness in AI: Addressing Gender Disparities in Machine Learning Models

Publication Date

1-1-2026

Document Type

Contribution to a Book

Publication Title

Smart Technologies for Sustainable Development Goals Gender Equality

DOI

10.1201/9781003685081-12

First Page

156

Last Page

168

Abstract

As artificial intelligence (AI) and machine learning (ML) get increasingly integrated into diverse sectors, the presence of gender bias in AI systems has emerged as a critical concern with far-reaching implications. In this chapter we explore manifestation, causes, measurement, and mitigation of gender bias in AI-ML applications. Some popular real-world hiring algorithms, facial recognition tools, AI-powered voice assistants, and popular applications, like Lensa, demonstrate the tangible and harmful impacts of biased AI systems. These biases typically stem from skewed training data, underrepresentation of certain gender groups, algorithmic design flaws, and biased feature selection processes. After pointing out the limitations of traditional AI performance metrics like accuracy, precision, recall, and F1 score, which may obscure underlying biases, we emphasize the importance of incorporating fairness-specific metrics such as statistical parity, equal opportunity, equality of odds, predictive parity, and treatment equality. Subsequently, in order to address this gender bias in AI, we examine a range of bias mitigation strategies, including the use of diverse and balanced datasets, data augmentation techniques, careful preprocessing and routine model testing. Through a comprehensive review and practical recommendations, this chapter underscores the urgent need for inclusive and responsible AI development practices to ensure fairness and equity in AI-driven decision-making.

Department

Computer Science

Link to Full Text

Share

COinS