Mitigating Racial Algorithmic Bias in Healthcare Artificial Intelligent Systems: A Fairness-Aware Machine Learning Approach

Racial bias in artificial intelligent (AI) systems is a pervasive problem that undermines fairness and perpetuates inequality. This paper presents an analysis of bias-mitigating algorithms applied to standard machine learning classifiers with the goal of reducing racial bias in healthcare access. The study integrates Reweighing, Adversarial Debiasing, and Exponentiated Gradient (with demographic parity) methods with two base classifiers: Logistic Regression and Random Forest Classifier. The evaluation framework employs both traditional predictive performance metrics—accuracy, precision, recall, and F1 score—and fairness metrics, including average odds difference, disparate impact score, equal opportunity difference, and balanced accuracy. The study finds that after the application of bias-mitigating algorithms, the Random Forest Classifier augmented with the Reweighing algorithm is the superior model, achieving an accuracy of 88.01%, precision of 89%, recall of 97%, and F1 score of 93%, while also demonstrating a balance with respect to fairness. This indicates the model's capability to maintain equity across diverse racial groups and the Random Forest Classifier model's compatibility with the Reweighing technique. The model was then interpreted using Local Interpretable Model-Agnostic Explanations to offer clear insights into the decision-making process. By highlighting effective strategies for mitigating racial algorithmic bias and fostering interpretability, this paper contributes foundational knowledge crucial for ethical AI development in the healthcare sector.