Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Data-Driven Analysis of Intersectional Bias in Image Classification: A Framework with Bias-Weighted Augmentation

Farjana Yesmin

TL;DR

Addresses intersectional biases in image classification by introducing the Intersectional Fairness Evaluation Framework (IFEF), which combines fairness metrics with interpretability analyses to diagnose bias patterns. It then proposes Bias-Weighted Augmentation (BWA), a data-driven augmentation strategy that scales transformations by subgroup underrepresentation to improve underrepresented intersections without harming overall accuracy. Empirical evaluation on Open Images V7 across five object classes shows that BWA reduces disparity in Demographic Parity and Equal Opportunity and yields substantial gains in underrepresented intersections (up to about 24 percentage points) with significance across multiple runs ($p < 0.05$). The work offers a transparent, reproducible framework for diagnosing and mitigating complex intersectional biases in visual recognition.

Abstract

Machine learning models trained on imbalanced datasets often exhibit intersectional biases-systematic errors arising from the interaction of multiple attributes such as object class and environmental conditions. This paper presents a data-driven framework for analyzing and mitigating such biases in image classification. We introduce the Intersectional Fairness Evaluation Framework (IFEF), which combines quantitative fairness metrics with interpretability tools to systematically identify bias patterns in model predictions. Building on this analysis, we propose Bias-Weighted Augmentation (BWA), a novel data augmentation strategy that adapts transformation intensities based on subgroup distribution statistics. Experiments on the Open Images V7 dataset with five object classes demonstrate that BWA improves accuracy for underrepresented class-environment intersections by up to 24 percentage points while reducing fairness metric disparities by 35%. Statistical analysis across multiple independent runs confirms the significance of improvements (p < 0.05). Our methodology provides a replicable approach for analyzing and addressing intersectional biases in image classification systems.

Data-Driven Analysis of Intersectional Bias in Image Classification: A Framework with Bias-Weighted Augmentation

TL;DR

Addresses intersectional biases in image classification by introducing the Intersectional Fairness Evaluation Framework (IFEF), which combines fairness metrics with interpretability analyses to diagnose bias patterns. It then proposes Bias-Weighted Augmentation (BWA), a data-driven augmentation strategy that scales transformations by subgroup underrepresentation to improve underrepresented intersections without harming overall accuracy. Empirical evaluation on Open Images V7 across five object classes shows that BWA reduces disparity in Demographic Parity and Equal Opportunity and yields substantial gains in underrepresented intersections (up to about 24 percentage points) with significance across multiple runs (). The work offers a transparent, reproducible framework for diagnosing and mitigating complex intersectional biases in visual recognition.

Abstract

Machine learning models trained on imbalanced datasets often exhibit intersectional biases-systematic errors arising from the interaction of multiple attributes such as object class and environmental conditions. This paper presents a data-driven framework for analyzing and mitigating such biases in image classification. We introduce the Intersectional Fairness Evaluation Framework (IFEF), which combines quantitative fairness metrics with interpretability tools to systematically identify bias patterns in model predictions. Building on this analysis, we propose Bias-Weighted Augmentation (BWA), a novel data augmentation strategy that adapts transformation intensities based on subgroup distribution statistics. Experiments on the Open Images V7 dataset with five object classes demonstrate that BWA improves accuracy for underrepresented class-environment intersections by up to 24 percentage points while reducing fairness metric disparities by 35%. Statistical analysis across multiple independent runs confirms the significance of improvements (p < 0.05). Our methodology provides a replicable approach for analyzing and addressing intersectional biases in image classification systems.
Paper Structure (33 sections, 4 equations, 5 figures, 4 tables, 1 algorithm)

This paper contains 33 sections, 4 equations, 5 figures, 4 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Research Problem
  3. Contributions
  4. Related Work
  5. Fairness in Machine Learning
  6. Data Augmentation for Bias Mitigation
  7. Interpretability for Fairness Analysis
  8. Methodology
  9. Problem Formalization
  10. Dataset and Environmental Attribute Extraction
  11. Intersectional Fairness Evaluation Framework (IFEF)
  12. Fairness Metrics.
  13. Performance Metrics.
  14. Interpretability Analysis.
  15. Bias-Weighted Augmentation (BWA)
  16. ...and 18 more sections

Figures (5)

  • Figure 1: Training set class distribution showing imbalance across object categories.
  • Figure 2: Environmental attribute analysis: (a) Lighting distribution by class, (b) Background complexity by class, (c) Lighting score distribution, (d) Complexity score distribution.
  • Figure 3: Performance comparison across models: (left) Per-class accuracy showing improvements for all classes, with largest gains for underrepresented classes, (right) F1-Score comparison demonstrating balanced precision-recall improvements.
  • Figure 4: Fairness metrics comparison: (left) Demographic Parity by class with disparity annotations showing reduction from 0.142 to 0.092, (right) Equal Opportunity showing reduced TPR disparities across classes from 0.187 to 0.121.
  • Figure 5: Confusion matrices: (left) Baseline model showing higher confusion between underrepresented classes, (right) BWA model demonstrating reduced misclassification rates and sharper diagonal.