Promoting Equity: Assessing Algorithmic Fairness in Machine Learning Approaches for Predicting Alzheimer's Disease

Alzheimer's disease (AD) presents a significant challenge for older adults in marginalized communities. The application of machine learning (ML) techniques through predictive models holds promise for enhancing the early detection and management of AD. However, it is essential to address potential biases within ML models that may contribute to or exacerbate existing disparities. This study delves into the algorithmic fairness of four distinct ML models, logistic regression (LR), random forest (RF), k-nearest neighbors (KNN), and multilayer perceptron (MLP), in predicting the Alzheimer's disease. The assessment of fairness spans race, ethnicity, and gender subgroups, employing major key measurements: demographic parity, equalized odds, and disparate impact. While all four ML models exhibited strong overall performance, disparities emerged across race and ethnicity subgroups. Participants with Hispanic ethnicity background experienced lower sensitivity compared to their Non-Hispanic counterparts. Similar patterns of decreased sensitivity were observed for non-white participants in contrast to Non-Hispanic White individuals. Prediction performance dropped when demographic features was included. Despite their aggregate accuracy, the ML models fell short in meeting fairness metrics, emphasizing the need to integrate fairness considerations in the phase ML models for AD prediction. Addressing these disparities is crucial for ensuring equitable outcomes in the utilization of predictive models within diverse populations.