Bias-Audit

Credit Card Approval – Bias & Fairness Audit

• 🌐 View Live Demo

This project conducts a fairness and bias audit on a credit card approval dataset. The main objective is to assess whether sensitive attributes such as Gender, Ethnicity, and Citizenship unfairly influence approval outcomes, evaluate fairness metrics, apply bias mitigation strategies, and provide recommendations aligned with ethical AI practices.

---

Project Objectives

• Audit the dataset for potential bias and discrimination.
• Evaluate fairness using multiple fairness metrics.
• Apply bias mitigation techniques.
• Provide recommendations for ethical AI adoption.

---

Dataset

• Source: Kaggle – Credit Card Approval Clean Data
• Sensitive Attributes: Gender, Ethnicity, Citizenship
• Target Variable: Approval (1 = approved, 0 = not approved)

---

Methods

1. Data Cleaning & Preprocessing

• Standardized numerical and categorical features.
• Isolated sensitive attributes for fairness checks.

2. Bias Analysis

• Compared approval rates across groups.
• Identified disparities in outcomes.

3. Fairness Metrics

• Demographic Parity
• Equal Opportunity
• Disparate Impact Ratio

4. Bias Mitigation Techniques

• Reweighing
• Pre-processing transformations
• Fairness-aware classifiers

---

Findings

Bias Analysis

• Gender: Males had higher approval rates compared to females.
• Ethnicity: Certain ethnic groups were disadvantaged with lower approval rates.
• Citizenship: Non-citizens had significantly lower approval rates.

Fairness Metrics

Metric	Finding	Bias Detected
Demographic Parity	Approval rates unequal across groups	Yes
Equal Opportunity	Qualified applicants treated unequally	Yes
Disparate Impact	Females & non-citizens < 0.8 ratio threshold	Yes

Bias Mitigation Results

• Disparities reduced after applying reweighing and fairness-aware models.
• Fairness metrics showed measurable improvement.

---

Recommendations

• Use fairness-aware algorithms in production systems.
• Conduct regular audits with fairness metrics.
• Collect more representative datasets to reduce systemic bias.
• Implement explainability techniques for transparency.
• Adopt ethical AI guidelines in financial decision-making.

---

Conclusion

The audit revealed systematic bias against women, certain ethnic groups, and non-citizens.
While mitigation techniques improved fairness, long-term solutions require better data practices, continuous monitoring, and ethical AI adoption.

This project demonstrates the importance of responsible AI in credit and financial systems.