Exploring Geriatric Clinical Data and Mitigating Bias with Multi-Objective Synthetic Data Generation for Equitable Health Predictions

In the data-centric landscape of modern healthcare, addressing bias in machine learning models is crucial for ensuring equitable health outcomes.When applied in clinical settings, biased predictions can exacerbate disparities in healthcare.This paper focuses on the domain of biomedical informatics and the challenge of mitigating bias in smart home datasets used for health monitoring.We assess existing bias metrics and a new metric, the Objective Fairness Index (OFI), to quantify bias related to sensitive attributes.To address these biases, we propose a novel method using a multiobjective generative adversarial network (GAN) that generates diverse synthetic data to improve data representation.This approach, validated on data from older adults managing chronic health conditions, demonstrates the potential to enhance both prediction accuracy and fairness in health outcomes.