2025 SOA ImpACT Conference: Session 1J: Bias, Fairness and the Bottom Line: Identifying and Mitigating Model Bias in Insurance-11.03.25 Recording

As predictive modeling and machine learning become increasingly embedded in insurance practices, the risk of unintended bias in models has become a pressing concern. Bias-whether introduced through data, assumptions, or algorithms-can impact fairness, regulatory compliance, and even profitability. This session will explore the multifaceted nature of model bias in insurance, examining its sources, implications, and strategies for mitigation. Speakers will share practical frameworks for identifying bias, assessing model performance across subgroups, and aligning modeling practices with ethical and regulatory standards. Designed for actuaries at all levels, from beginners looking to understand foundational concepts to experts seeking advanced mitigation techniques, this session will provide actionable insights into building more transparent, equitable, and trustworthy models. Key takeaways: - Define different types of model bias and their implications in insurance - Understand methods for detecting and quantifying bias in models - Learn practical approaches to mitigating bias while preserving model performance - Explore ethical and regulatory considerations shaping the future of actuarial modeling