Key result
Balancing the dataset through undersampling improved the Multilayer Perceptron F1-score for predicting diabetes from 0.38 to 0.78 and mitigated gender bias.
Why the study?
Type 2 diabetes prevalence is increasing in young adults, but machine learning models exhibit bias against this minority age group and lack effective predictive methodologies.
Does balancing datasets via undersampling improve the accuracy and fairness of machine learning models in predicting diabetes and prediabetes?
Cross-Sectional (n=441,456)
Does balancing datasets via undersampling improve the accuracy and fairness of machine learning models in predicting diabetes and prediabetes?
Undersampling techniques to balance training data significantly improve both the accuracy and gender fairness of machine learning models predicting diabetes.
This is the authors' abstract. We don't add key points for this paper.
While type 2 diabetes is predominantly found in the elderly population, recent publications indicate an increasing prevalence in the young adult population. Failing to predict it in the minority younger age group could have significant adverse effects on their health. The previous work acknowledges the bias of machine learning models towards different gender and race groups and proposes various approaches to mitigate it. However, prior work has not proposed any effective methodologies to predict diabetes in the young population which is the minority group in the diabetic population. In this paper, we identify this deficiency in traditional machine learning models and implement double prioritization (DP) bias correction techniques to mitigate the bias towards the young population when predicting diabetes. Deviating from the traditional concept of one-model-fits-all, we train customized machine-learning models for each age group. The DP model consistently improves recall of diabetes class by 26 to 40 in the young age group (30-44). Moreover, the DP technique outperforms 7 commonly used whole-group sampling techniques such as random oversampling, SMOTE, and AdaSyns techniques by at least 36 in terms of diabetes recall in the young age group. We also analyze the feature importance to investigate the source of bias in the original model. Our code is publicly available at an anonymous reposi- tory - https://anonymous.4open.science/r/Diabetes-BRFSS- DP-C847
May support undersampling in diabetes ML models; leaves open translation to clinical outcomes.
No takes yet. Share an insight, caveat, or question.
Pias et al. (2023) conducted a cross-sectional in Type 2 Diabetes and Prediabetes (n=441,456). Undersampling to balance dataset vs. Original imbalanced dataset was evaluated on F1-score of Multilayer Perceptron for predicting diabetes. Balancing the dataset through undersampling improved the Multilayer Perceptron F1-score for predicting diabetes from 0.38 to 0.78 and mitigated gender bias.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: