Fairness and bias correction in machine learning for depression prediction across four study populations
Abstract A significant level of stigma and inequality exists in mental healthcare, especially in under-served populations. Inequalities are reflected in the data collected for scientific purposes. When not properly accounted for, machine learning (ML) models learned from data can reinforce these str...
Main Authors: | , , , , , , |
---|---|
Format: | Article |
Language: | English |
Published: |
Nature Portfolio
2024-04-01
|
Series: | Scientific Reports |
Subjects: | |
Online Access: | https://doi.org/10.1038/s41598-024-58427-7 |
_version_ | 1797219861826371584 |
---|---|
author | Vien Ngoc Dang Anna Cascarano Rosa H. Mulder Charlotte Cecil Maria A. Zuluaga Jerónimo Hernández-González Karim Lekadir |
author_facet | Vien Ngoc Dang Anna Cascarano Rosa H. Mulder Charlotte Cecil Maria A. Zuluaga Jerónimo Hernández-González Karim Lekadir |
author_sort | Vien Ngoc Dang |
collection | DOAJ |
description | Abstract A significant level of stigma and inequality exists in mental healthcare, especially in under-served populations. Inequalities are reflected in the data collected for scientific purposes. When not properly accounted for, machine learning (ML) models learned from data can reinforce these structural inequalities or biases. Here, we present a systematic study of bias in ML models designed to predict depression in four different case studies covering different countries and populations. We find that standard ML approaches regularly present biased behaviors. We also show that mitigation techniques, both standard and our own post-hoc method, can be effective in reducing the level of unfair bias. There is no one best ML model for depression prediction that provides equality of outcomes. This emphasizes the importance of analyzing fairness during model selection and transparent reporting about the impact of debiasing interventions. Finally, we also identify positive habits and open challenges that practitioners could follow to enhance fairness in their models. |
first_indexed | 2024-04-24T12:40:23Z |
format | Article |
id | doaj.art-4ca8b1d4989c428790c5207bbdf5a179 |
institution | Directory Open Access Journal |
issn | 2045-2322 |
language | English |
last_indexed | 2024-04-24T12:40:23Z |
publishDate | 2024-04-01 |
publisher | Nature Portfolio |
record_format | Article |
series | Scientific Reports |
spelling | doaj.art-4ca8b1d4989c428790c5207bbdf5a1792024-04-07T11:16:38ZengNature PortfolioScientific Reports2045-23222024-04-0114111210.1038/s41598-024-58427-7Fairness and bias correction in machine learning for depression prediction across four study populationsVien Ngoc Dang0Anna Cascarano1Rosa H. Mulder2Charlotte Cecil3Maria A. Zuluaga4Jerónimo Hernández-González5Karim Lekadir6Departament de Matemàtiques i Informàtica, Facultat de Matemàtiques i Informàtica, Universitat de BarcelonaDepartament de Matemàtiques i Informàtica, Facultat de Matemàtiques i Informàtica, Universitat de BarcelonaDepartment of Child and Adolescent Psychiatry/Psychology, Erasmus MC, University Medical Center RotterdamDepartment of Child and Adolescent Psychiatry/Psychology, Erasmus MC, University Medical Center RotterdamData Science Department, EURECOMDepartament d’Informàtica, Matemàtica Aplicada i Estadística, Universitat de GironaDepartament de Matemàtiques i Informàtica, Facultat de Matemàtiques i Informàtica, Universitat de BarcelonaAbstract A significant level of stigma and inequality exists in mental healthcare, especially in under-served populations. Inequalities are reflected in the data collected for scientific purposes. When not properly accounted for, machine learning (ML) models learned from data can reinforce these structural inequalities or biases. Here, we present a systematic study of bias in ML models designed to predict depression in four different case studies covering different countries and populations. We find that standard ML approaches regularly present biased behaviors. We also show that mitigation techniques, both standard and our own post-hoc method, can be effective in reducing the level of unfair bias. There is no one best ML model for depression prediction that provides equality of outcomes. This emphasizes the importance of analyzing fairness during model selection and transparent reporting about the impact of debiasing interventions. Finally, we also identify positive habits and open challenges that practitioners could follow to enhance fairness in their models.https://doi.org/10.1038/s41598-024-58427-7Machine learning for depression predictionAlgorithmic fairnessBias mitigationNovel post-hoc methodPsychiatric healthcare equity |
spellingShingle | Vien Ngoc Dang Anna Cascarano Rosa H. Mulder Charlotte Cecil Maria A. Zuluaga Jerónimo Hernández-González Karim Lekadir Fairness and bias correction in machine learning for depression prediction across four study populations Scientific Reports Machine learning for depression prediction Algorithmic fairness Bias mitigation Novel post-hoc method Psychiatric healthcare equity |
title | Fairness and bias correction in machine learning for depression prediction across four study populations |
title_full | Fairness and bias correction in machine learning for depression prediction across four study populations |
title_fullStr | Fairness and bias correction in machine learning for depression prediction across four study populations |
title_full_unstemmed | Fairness and bias correction in machine learning for depression prediction across four study populations |
title_short | Fairness and bias correction in machine learning for depression prediction across four study populations |
title_sort | fairness and bias correction in machine learning for depression prediction across four study populations |
topic | Machine learning for depression prediction Algorithmic fairness Bias mitigation Novel post-hoc method Psychiatric healthcare equity |
url | https://doi.org/10.1038/s41598-024-58427-7 |
work_keys_str_mv | AT vienngocdang fairnessandbiascorrectioninmachinelearningfordepressionpredictionacrossfourstudypopulations AT annacascarano fairnessandbiascorrectioninmachinelearningfordepressionpredictionacrossfourstudypopulations AT rosahmulder fairnessandbiascorrectioninmachinelearningfordepressionpredictionacrossfourstudypopulations AT charlottececil fairnessandbiascorrectioninmachinelearningfordepressionpredictionacrossfourstudypopulations AT mariaazuluaga fairnessandbiascorrectioninmachinelearningfordepressionpredictionacrossfourstudypopulations AT jeronimohernandezgonzalez fairnessandbiascorrectioninmachinelearningfordepressionpredictionacrossfourstudypopulations AT karimlekadir fairnessandbiascorrectioninmachinelearningfordepressionpredictionacrossfourstudypopulations |