Scholarlink

This study presents a fairness-aware machine learning framework for biomedical prediction, illustrated through gallstone disease detection. Using a structured clinical dataset of 320 patients, predictive models, including multilayer perceptron and random forest, are developed via stratified 5-fold cross-validation. Model performance is evaluated using AUC, F1-score, precision, and recall, with subgroup fairness assessed through demographic parity, disparate impact, predictive parity difference, equal opportunity, and equalized odds. Bias linked to age and gender is quantified and corrected using threshold optimization and F1-based recalibration. Postprocessing interventions demonstrate improved fairness metrics while preserving diagnostic accuracy. Our methodology aligns with ethical guidelines and technical expectations under the EU AI Act, emphasizing equitable model behavior in clinical decision-making. The proposed framework contributes to responsible AI deployment in healthcare, supporting inclusion and transparency in high-risk biomedical applications.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Fairness-Aware Machine Learning for Biomedical Prediction: Evaluating and Correcting Bias in Gallstone Diagnosis Models

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Caroline König,
  • Martha Ivon Cardenas,
  • Pedro Jesús Copado,
  • Alfredo Vellido

摘要

This study presents a fairness-aware machine learning framework for biomedical prediction, illustrated through gallstone disease detection. Using a structured clinical dataset of 320 patients, predictive models, including multilayer perceptron and random forest, are developed via stratified 5-fold cross-validation. Model performance is evaluated using AUC, F1-score, precision, and recall, with subgroup fairness assessed through demographic parity, disparate impact, predictive parity difference, equal opportunity, and equalized odds. Bias linked to age and gender is quantified and corrected using threshold optimization and F1-based recalibration. Postprocessing interventions demonstrate improved fairness metrics while preserving diagnostic accuracy. Our methodology aligns with ethical guidelines and technical expectations under the EU AI Act, emphasizing equitable model behavior in clinical decision-making. The proposed framework contributes to responsible AI deployment in healthcare, supporting inclusion and transparency in high-risk biomedical applications.