Code for ML4H'22 submission
This repository contains the code used in the project investigating algorithmic bias in delirium predicting machine learning models.
We used two datasets: ACTFAST Dataset (Epic era data from Barnes Jewish Hopital in St Louis; not available publicly) and MIMIC-III.
For the MIMIC III dataset, the extraction steps were followed from the supplementary information available in the paper on labelling delirum.
The psql queries for MIMICIII extraction can be found in MIMIC3-delirium_extraction.txt file.
For processing the datasets, partitioning into train-test and training the RF (others models were also tried), we used the file Delirium_model_fairness_testing.py and Delirium_model_fairness_testing_MIMIC.py for ACTFAST (Epic) and MIMIC-III respectively.
For computing the groupwise performance as reported in the paper, we used Differential_perf_generator.py and Differential_per_generator_MIMIC.py.
Dependencies: Python 3.9.10