Super2021

Team: Super2021

Project: BeFair (A Bias Detection Tool)

Team Members: Yinghao Zhu, Jingkun An, Enshen Zhou, Hao Li, Haoran Feng

We are students from Beihang University.

Research Interests: deep-based clinical predictive modeling with electronic health record (EHR) data

Brief Introduction

We designed an adaptive threshold method to evaluate the level of bias in both the dataset and the model, and implemented a multi-level reweighting method to correct for bias. We validated our approach on clinical dataset (MEPSDataset19) and adult census income dataset (AdultDataset). The proposed method found significant performance improvements.

Link to Video: BeFair: A Multi-Level-Reweighing Method to Mitigate Bias for Healthcare (YouTube)

Github Repository

Contains required measure_disparity.py, mitigate_disparity.py and Jupyter notebooks example_{dataset}.ipynb that tells how to use our proposed methods (correspond to above two Python file) on two datasets.

Github Repository Link

Supporting Documentation

Including following topics: Methodology Overview, Value Proposition, Healthcare Scenario, Operational Requirements, Sustainability Plan, Generalizability Plan, Implementation Requirements, Lessons Learned

Supporting Documentation PDF Link