Conference proceeding
A Collaborative Learning Approach for Fairness in Prediction of Substance Use Disorder Treatment Completion
Artificial Intelligence in Medicine, v 15735, pp 235-239
22 Jun 2025
Abstract
The use of machine learning for predictive modeling in healthcare provides great opportunities for knowledge discovery and decision support for clinicians. However, unequal performance of models across demographic groups can lead to disparities in health outcomes, potentially causing harm. Many approaches have been proposed to reduce the unfairness of machine learning models, and for fair model development, but these approaches may not consider or leverage the underlying differences between groups. The objective of this study was to utilize a collaborative learning approach to combine knowledge from models trained on the individual demographic groups. In the case of race groups, where the data was very imbalanced (10% Non Caucasian), collaborative learning effectively improved model fairness from the baseline with a drop in equalized odds (0.091 to 0.049) and a small drop in performance (AUC 0.846 to 0.823). Using sex groups for collaborative learning, where the distribution was not as imbalanced (35% female), did not lead to significant changes, suggesting that the collaborative approach has potential for developing a fair model in cases where the sensitive attribute distribution is heavily imbalanced.
Metrics
3 Record Views
Details
- Title
- A Collaborative Learning Approach for Fairness in Prediction of Substance Use Disorder Treatment Completion
- Creators
- Mary M. LucasChristopher C. Yang
- Contributors
- Riccardo Bellazzi (Editor)José Manuel Juarez Herrero (Editor)Lucia Sacchi (Editor)Blaž Zupan (Editor)
- Publication Details
- Artificial Intelligence in Medicine, v 15735, pp 235-239
- Series
- Lecture Notes in Computer Science
- Publisher
- Springer Nature ; Cham
- Number of pages
- 5
- Grant note
- National Science FoundationDepartment of Defense: DoD W91XWH-05-1-023
This work was supported in part by the National Science Foundation under the Grants IIS-1741306 and IIS-2235548, and by the Department of Defense under the Grant DoD W91XWH-05-1-023. This material is based upon work supported by (while serving at) the National Science Foundation. Any opinions, findings, and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the National Science Foundation.
- Resource Type
- Conference proceeding
- Language
- English
- Academic Unit
- Information Science (Informatics)
- Web of Science ID
- WOS:001553213900042
- Scopus ID
- 2-s2.0-105009824751
- Other Identifier
- 9783031958403; 991022059740404721