biasFairnessDemo

A compact, self-contained demonstration notebook for bias and fairness analysis in supervised machine learning.

This repository provides a hands-on workflow for:

• training a baseline classifier,
• inspecting performance across sensitive groups,
• computing fairness indicators,
• and exploring how different modeling choices affect accuracy–fairness trade-offs.

The demo is implemented as a single Jupyter notebook and uses the Adult Income dataset.

---

What this demo illustrates

Bias and fairness issues can appear even when a model performs well globally. Typical patterns include:

• systematic performance differences between demographic groups (e.g., higher false negative rates for one group),
• disparate rates of positive predictions (disparate impact),
• indirect discrimination caused by correlated features (proxy variables).

This notebook guides through a fairness audit in a transparent, didactic way:

1. Train a baseline model
2. Measure group-wise outcomes
3. Quantify fairness criteria
4. Discuss trade-offs and mitigation strategies

---

Contents

• biasFairnessDemo.ipynb
  Main notebook implementing the full demo pipeline.

• adult_income_dataset.csv
  Input dataset used in the notebook (preprocessed/exported variant of the Adult dataset).

• figures/
  Generated plots/illustrations used in the notebook or exported for presentations.

• requirements.txt
  Python dependencies used for execution.

---

Repository structure

biasFairnessDemo/
├── figures/                    # plots used/generated by the demo
├── adult_income_dataset.csv    # dataset for the fairness demo
├── biasFairnessDemo.ipynb      # main notebook (bias & fairness analysis workflow)
├── requirements.txt
└── README.md

---

Installation

1. Clone the repository

git clone https://github.com/stefan-ehrlich/biasFairnessDemo.git
cd biasFairnessDemo  

2. Create python environment

conda create -n biasEnv python=3.10
conda activate biasEnv

3. Install dependencies

pip install -r requirements.txt

4. Run notebook

jupyter notebook