Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Evaluating Pre-Training Bias on Severe Acute Respiratory Syndrome Dataset

Diego Dimer Rodrigues

TL;DR

This work investigates pre-training bias in a Brazilian health dataset (SRAG) by visualizing three bias metrics—Class Imbalance, Kullback-Leibler Divergence, and Kolmogorov-Smirnov—across five regions. By training region-specific random forests and evaluating cross-region performance, it assesses whether regional data differences translate into biased predictions for protected attributes like sex and race. Across 2021–2023, the study generally finds low bias signals from KL and KS, with occasional CI hints of imbalance, and demonstrates that visualization combined with cross-region evaluation can help detect and interpret potential harms before model training. The findings suggest the dataset and public health context may support fair predictions, though the authors emphasize using multiple metrics to capture different facets of bias and the importance of visualization for bias mitigation.

Abstract

Machine learning (ML) is a growing field of computer science that has found many practical applications in several domains, including Health. However, as data grows in size and availability, and the number of models that aim to aid or replace human decisions, it raises the concern that these models can be susceptible to bias, which can lead to harm to specific individuals by basing its decisions on protected attributes such as gender, religion, sexual orientation, ethnicity, and others. Visualization techniques might generate insights and help summarize large datasets, enabling data scientists to understand the data better before training a model by evaluating pre-training metrics applied to the datasets before training, which might contribute to identifying potential harm before any effort is put into training and deploying the models. This work uses the severe acute respiratory syndrome dataset from OpenDataSUS to visualize three pre-training bias metrics and their distribution across different regions in Brazil. A random forest model is trained in each region and applied to the others. The aim is to compare the bias for the different regions, focusing on their protected attributes and comparing the model's performance with the metric values.

Evaluating Pre-Training Bias on Severe Acute Respiratory Syndrome Dataset

TL;DR

This work investigates pre-training bias in a Brazilian health dataset (SRAG) by visualizing three bias metrics—Class Imbalance, Kullback-Leibler Divergence, and Kolmogorov-Smirnov—across five regions. By training region-specific random forests and evaluating cross-region performance, it assesses whether regional data differences translate into biased predictions for protected attributes like sex and race. Across 2021–2023, the study generally finds low bias signals from KL and KS, with occasional CI hints of imbalance, and demonstrates that visualization combined with cross-region evaluation can help detect and interpret potential harms before model training. The findings suggest the dataset and public health context may support fair predictions, though the authors emphasize using multiple metrics to capture different facets of bias and the importance of visualization for bias mitigation.

Abstract

Machine learning (ML) is a growing field of computer science that has found many practical applications in several domains, including Health. However, as data grows in size and availability, and the number of models that aim to aid or replace human decisions, it raises the concern that these models can be susceptible to bias, which can lead to harm to specific individuals by basing its decisions on protected attributes such as gender, religion, sexual orientation, ethnicity, and others. Visualization techniques might generate insights and help summarize large datasets, enabling data scientists to understand the data better before training a model by evaluating pre-training metrics applied to the datasets before training, which might contribute to identifying potential harm before any effort is put into training and deploying the models. This work uses the severe acute respiratory syndrome dataset from OpenDataSUS to visualize three pre-training bias metrics and their distribution across different regions in Brazil. A random forest model is trained in each region and applied to the others. The aim is to compare the bias for the different regions, focusing on their protected attributes and comparing the model's performance with the metric values.
Paper Structure (16 sections, 3 equations, 9 figures)

This paper contains 16 sections, 3 equations, 9 figures.

Table of Contents

  1. Introduction
  2. Theoretical Background
  3. Bias in Machine Learning
  4. Pre-training Metrics
  5. Class Imbalance
  6. Kullback-Leibler (KL) Divergence
  7. Kolmogorov-Smirnov (KS)
  8. Related Work
  9. Methodology
  10. Data Collection
  11. Training and Performance Evaluation
  12. Results
  13. 2021
  14. 2022
  15. 2023
  16. ...and 1 more sections

Figures (9)

  • Figure 1: Choropleth maps for 2021, for Sex (right) and Race (left)
  • Figure 2: Prediction analysis for region north using a model trained on the southeast, for 2021
  • Figure 3: Prediction analysis for region south using a model trained on the north, for 2021
  • Figure 4: Choropleth maps for 2022, for Sex (right) and Race (left)
  • Figure 5: Prediction analysis for region midwest using a model trained on the southeast, for 2022
  • ...and 4 more figures