Project-Nova 🤖

Introduction

Project-Nova is a project that demonstrates the process of identifying and mitigating bias in a machine learning model. The project focuses on a scenario of predicting the creditworthiness of partners, where the initial dataset contains inherent biases based on gender and geographic location.

This notebook walks through the following key steps:

1. Generation of a biased dataset: A synthetic dataset is created with intentional biases to simulate a real-world scenario.
2. Bias detection and analysis: The biases in the dataset are verified and analyzed.
3. Model training: A predictive model is trained on the biased dataset.
4. Bias mitigation: A mitigation technique is applied to the model's predictions to ensure fairness.
5. Evaluation of mitigation impact: The effect of the bias mitigation is analyzed and visualized.

---

✨ Features

• Synthetic Data Generation: Generates a custom dataset with controllable biases.
• Bias Verification: Provides statistical verification of the introduced biases.
• Creditworthiness Prediction: Trains a model to predict whether a partner is creditworthy.
• Bias Mitigation: Implements a post-processing technique to mitigate bias in the model's predictions.
• Impact Visualization: Visualizes the impact of the mitigation process, showing how many partners' classifications were changed to achieve fairness.

---

🚀 How to Run

1. Clone the repository:

   git clone [https://github.com/Samridh-Minocha007/Project-Nova.git](https://github.com/Samridh-Minocha007/Project-Nova.git)

2. Navigate to the project directory:

   cd Project-Nova

3. Install the required dependencies:

   pip install pandas numpy scikit-learn matplotlib seaborn

4. Open and run the Jupyter Notebook:

   jupyter notebook nova_prototype_new.ipynb

---

⚙️ Dependencies

The project requires the following Python libraries:

• pandas
• numpy
• scikit-learn
• matplotlib
• seaborn

You can install all the dependencies by running the command in the "How to Run" section.

---

📊 Results

The notebook generates several outputs, including:

• A biased dataset saved as nova_custom_biased_dataset.csv.
• Verification of biases in the dataset, showing disparities in earnings and GMV based on gender and location.
• A visualization of the impact of bias mitigation in the form of a movement matrix, which shows how many partners were upgraded or downgraded to ensure fairness.

Here is an example of the mitigation impact matrix generated by the notebook:

This matrix clearly shows how many partners were moved from "Not Creditworthy" to "Creditworthy" and vice-versa after applying the bias mitigation technique.