Evaluating the Performance Disparity and the Role of Gender-Aware Approaches in Machine Learning Based Disease Detection

Abstract

Machine Learning (ML) is gaining attraction in medical research due to its ability to identify unnoticeable patterns by the human eye. However, concerns about fairness in ML models, particularly performance differences across groups, are growing. This study, therefore, focuses on evaluating the performance disparity and the role of gender-aware approaches in ML-based disease detection. It uses the gender-aware approach and introduces its two new variants by testing them on nine different disease datasets. Intensive experimental evaluations reveal that the detection performance can increase up to an F1-score of 1.0, depending on the nature of the dataset at hand. On the other hand, the genderaware approach is successful in mitigating the performance disparity only in three out of nine cases. The variants relying on a crossing-over fashion can capture the relationships and different patterns in some cases, but often fall behind the gender-aware approach. This research distinguishes itself through the use of a significant number of datasets and implemented pipelines, of which two are employed for mitigating performance disparity in disease detection for the first time in the literature. The findings of this study, therefore, make important contributions to the field of disease detection in terms of the aforementioned aspects.