Bias Detection and Mitigation in HR Algorithms: Ensuring Fairness in Machine Learning Models

Abstract

Listen

Translate article

Additional efforts are necessary to guarantee that the machine learning algorithms employed in healthcare do not perpetuate or exacerbate any preexisting discriminatory or objectionable biases that may be present in the data. In order to reduce the impact of any biases that may have developed during the data collection procedure, we implement a reinforcement learning approach. We assessed the effectiveness of our model in accurately predicting the incidence of COVID-19 in patients seeking medical attention at hospital emergency departments in order to reduce the impact of potential biases associated with ethnicity and hospital location. By utilizing an innovative incentive function and training strategy, we have proven that our methodology outperforms the most advanced machine learning techniques in terms of clinically significant screening outcomes and equitable results. To illustrate the adaptability of our methodology, we implemented a comprehensive evaluation of our approach in an intensive care unit by administering a discharge status test. Additionally, we conducted external validation at three distinct institutions.