Reading and Implementation on Data Bias

For this assignment, please read the following book chapters and articles:
1. Gender Shades: Intersectional Accuracy Disparities in Commercial Gender Classification, Joy Buolamwini and Timnit Gebru, PMLR, 2018
2. Bias Mitigation for Machine Learning Classifiers: A Comprehensive Survey, Max Hort et al., ACM Journal on Responsible Computing, 2024
3. Assessing and Remedying Coverage for a Given Dataset, Abolfazl Asudeh et al., IEEE ICDE, 2019
4. AI Fairness 360: An Extensible Toolkit for Detecting, Understanding, and Mitigating Unwanted Algorithmic Bias, Rachel KE Bellamy et al., arXiv preprint, 2018
5. Practical Fairness, Aileen Nielsen (ch. 3-4)
6. Machine Bias, Julia Angwin et al., ProPublica, 2016 [link]
7. AI Fairness 360 [link]
8. Data Preprocessing Techniques for Classification without Discrimination, Faisal Kamiran, Knowledge and Information Systems, 2012
9. Certifying and Removing Disparate Impact, Michael Feldman et al., ACM KDD, 2015
10. Learning Fair Representations, Zemel, Rich, et al., PMLR, 2013
After completing these readings, please critical answer the following questions.

Based on paper [1], answer the following questions:
1. What are the suggested solutions by the paper to mitigatge bias? What are the challenges of implementing such solutions in practice, particularly for commercial AI developers?
2. The study highlights the importance of intersectional evaluations. How can developers and researchers ensure that AI systems are rigorously tested across multiple demographic axes without oversimplifying complex identities?

Based on paper [2], answer the following questions:
Paper [2] considers four approaches for pre-processing-based bias mitigation (see Section 4), which are: (1) Relabeling and Perturbation, (2) Sampling, (3) Latent Variables, and (4) Representation. Please provide a detailed explanation of each of these approaches, focusing on how they work to mitigate bias in machine learning classifiers. Then, by defining a specific use case (e.g., hiring prediction, credit approval, healthcare diagnosis) explain how each approach could be applied. Explain the advantages and disadvantages of using each approach in your chosen use case (for example, by considering factors such as effectiveness in bias reduction, impact on data quality, scalability and complexity, etc.)

Based on papers [3-10], answer the following questions:
Using the concepts presented in paper [3], analyze and identify biases within the COMPAS dataset [6]. Your task is to evaluate coverage issues and fairness using the AIF360 library [4][5][7]. Specifically, you should:
1. Assess Coverage: Investigate the distribution of subpopulations based on protected attributes (e.g., race, gender) as well as the distribution of the outcome variable (recidivism) across these groups in the COMPAS dataset. Identify any underrepresented or missing subpopulations. Determine whether the distribution of favorable vs. unfavorable outcomes is balanced across protected attributes.
2. Measure Bias: Compute the following fairness metrics for race and gender: (i) statistical parity to measure the difference in favorable outcomes between privileged and unprivileged groups, (ii) predictive parity that assesses whether the probability of a correct positive prediction (i.e., predicting recidivism for those who reoffend) is the same for different groups, and (iii) equalized odds that requires both true positive rates (TPR) and false positive rates (FPR) be the same across groups. Compare and contrast the results. Which metric provides the most meaningful insight into bias? Do these metrics align, or do they reveal different aspects of bias?
3. Train a Classifier: Train a logistic regression model to predict recidivism using the dataset. After training the model, evaluate its predictions for bias using fairness metrics defined at step 2. Assess the trade-off between fairness and accuracy. How does the model perform in terms of accuracy for privileged vs. unprivileged groups? Is improving fairness always at odds with model performance? Does your model exhibit any bias across groups?
4. Apply Bias Mitigation Techniques: Apply the following pre-processing bias mitigation techniques to reduce bias in your model: (i) reweighting that adjusts the weights of individuals in the dataset to ensure that privileged and unprivileged groups receive equal consideration during model training [8], (ii) disparate impact remover that modifies feature values to reduce the impact of protected attributes (such as race and gender) on the predictions [9], and (iii) learning fair representations (LFR) that learns a new, fair feature representation of the data [10]. Compare the effectiveness of these techniques in reducing bias. Do these bias mitigation strategies lead to better outcomes and reduce bias across all groups?