Fairness-Aware Income Prediction using Machine Learning

This repository contains the implementation of a fairness-aware machine learning pipeline developed as part of a Master’s thesis project.
The project focuses on predicting income levels using the Adult Income Dataset while systematically analyzing algorithmic fairness across demographic groups.

📌 Project Objectives

• Build machine learning models to predict whether an individual earns more than $50K/year.
• Evaluate standard performance metrics (accuracy, confusion matrix).
• Perform group-specific fairness analysis across sensitive attributes (e.g., sex, race).
• Analyze how fairness metrics evolve across different models and datasets.
• Provide reproducible, research-quality results suitable for academic evaluation.

📊 Dataset

The project uses the Adult Income Dataset, which contains demographic and employment-related features such as:

• Age, education, occupation
• Gender, race
• Capital gain/loss
• Income label (<=50K, >50K)

Two versions of the dataset are used:

• Raw dataset (adult.csv)
• Preprocessed dataset (Adult pre-processed dataset.csv)

🧠 Models Implemented

• Logistic Regression
• Random Forest

Each model is evaluated on:

• Predictive performance
• Feature importance
• Group-level fairness metrics

⚖️ Fairness Evaluation

To assess fairness, the project computes group-specific metrics for protected attributes such as sex and race.

Group-Specific Metrics:

• Confusion Matrix (TN, FP, FN, TP)
• Base Rate
• Positive Prediction Rate
• True Positive Rate (TPR)
• False Positive Rate (FPR)
• Precision

These metrics help identify potential biases and evaluate fairness notions such as:

• Equal Opportunity
• Equalized Odds

Fairness results are saved as CSV files for transparency and further analysis.

📁 Project Structure

thesis_code/ │ ├── main.py # Entry point for running the full pipeline ├── config.py # Configuration for datasets and output paths ├── data_loader.py # Data loading and preprocessing ├── models.py # Model training and evaluation functions ├── evaluation.py # Performance and fairness evaluation logic ├── fairness.py # Group-specific fairness metrics ├── plot_utils.py # Confusion matrix and feature importance plots ├── visualizations.py # Fairness metric visualizations │ ├── results/ # CSV and JSON outputs └── figures/ # Generated plots

▶️ How to Run the Project

1. Install dependencies

pip install numpy pandas scikit-learn matplotlib seaborn

📈 Outputs
After execution, the following outputs are generated:
📂 results/
performance_metrics.csv
fairness_metrics.csv
Group-specific fairness reports per dataset and model

📂 figures/
Confusion matrix plots
Feature importance plots
All outputs are automatically saved for reproducibility.

🎓 Academic Context
This project was developed as part of a Master’s thesis focusing on:
Ethical AI
Bias and fairness in machine learning
Responsible data-driven decision-making
The implementation follows best practices for modularity, reproducibility, and interpretability.

📜 License

This project is intended for academic and research use.
Please cite appropriately if used in publications or derivative works.