Investigating the effect of bias mitigation in machine learning algorithms

Machine learning has been proven effective when applied to healthcare applications like patient diagnosis or hospital admission prediction. However, with these models comes the risk of algorithmic bias causing possible unfairness. This is associated with data distribution or lack of data accounting for a diverse population. The biggest differentiators identified in literature for algorithmic bias are race, gender and age. If not identified and mitigated properly, algorithmic bias in machine learning systems can drastically impact patient treatment outcomes for the worse. The effect of bias based on race and gender has been explored frequently in literature. While age bias has been examined in literature to some extent, to our knowledge, there is no study that investigates the utility of a fairness-based framework to detect and mitigate age bias. Hence, our paper examines the existence and mitigation of age bias in machine learning applications, using AI Fairness 360 framework within healthcare domain.