Data to Decision

This repository contains the code and data for the research project "From Data to Decision: Investigating Bias Amplification in Decision-Making Algorithms". The project examines how biases present in the Adult/Census Income dataset influence the outputs of machine learning models, exploring whether these biases are amplified by Logistic Regression, Decision Tree, and Random Forest algorithms using various fairness metrics.

• Title: From Data to Decision: Investigating Bias Amplification in Decision-Making Algorithms
• Date: June 23, 2024
• Institution: Delft University of Technology
• Link: Download the thesis (PDF)

---

Repository Structure

data-to-decision/
├── adult.csv
├── preprocess/
│   ├── preprocess_remove.py
│   ├── preprocess_impute_simple.py
│   ├── adult_preprocessed_train.csv
│   └── adult_preprocessed_test.csv
├── algorithms/
│   ├── algs.py
│   └── adult_trained_test_results.csv
├── bias/
│   ├── metrics_race.py
│   ├── metrics_sex.py
│   ├── metrics_plots.py
│   ├── race_plots.py
│   ├── sex_plots.py
│   └── dummy/
│       ├── data.csv
│       ├── pred.csv
│       └── metrics_dummy.py
└── README.md

• preprocess/: Scripts and preprocessed data for cleaning, encoding, and scaling the raw dataset. Note that preprocess_impute_simple.py is maintained for reference but should not be used in the analysis pipeline.
• bias/: Scripts to compute fairness metrics (Demographic Parity, Disparate Impact, Equal Opportunity, Equalized Odds) and generate corresponding plots.
• algorithms/: Contains algs.py, which trains Logistic Regression, Decision Tree, and Random Forest models on the preprocessed data, and outputs test set predictions and performance metrics to adult_trained_test_results.csv.

---

Installation

1. Clone the repository:

   git clone https://github.com/emihalache/data-to-decision.git
   cd data-to-decision

2. (Optional) Create a virtual environment:

   python3 -m venv venv
   source venv/bin/activate

3. Install dependencies:

   pip install pandas numpy scikit-learn fairlearn matplotlib seaborn

---

Usage

1. Data Preprocessing

   • Run the preprocessing pipeline to clean, encode, and impute missing values:

     python preprocess/preprocess_remove.py

   • The processed training and test sets will be saved as adult_preprocessed_train.csv and adult_preprocessed_test.csv in the preprocess/ directory.

2. Model Training

   • Train the classification models and generate test-set results:

     python algorithms/algs.py

   • The script will train Logistic Regression, Decision Tree, and Random Forest models, then save predictions and performance metrics to algorithms/adult_trained_test_results.csv.

3. Bias Analysis

   • Compute fairness metrics and generate plots:

     cd bias

   • Then run:

     python metrics_race.py
     python race_plots.py
     python metrics_sex.py
     python sex_plots.py

   • When finished, return to the root:

     cd ..

---

Requirements

• Python 3.7 or higher
• pandas
• numpy
• scikit-learn
• fairlearn
• matplotlib
• seaborn

---

Acknowledgements

• Adult/Census Income Dataset: UCI Machine Learning Repository

---