Quantum Machine Learning on Bankruptcy Prediction

This project explores the application of both classical and quantum machine learning techniques for predicting bankruptcy based on financial datasets. The analysis includes data preprocessing, feature selection, classical machine learning models, and Variational Quantum Classifiers (VQC).

Project Overview

• Dataset: Financial indicators for bankruptcy prediction.
• Objective: Compare the performance of classical machine learning models and quantum machine learning using Qiskit.
• Techniques Used:
  • Classical Models: K-Nearest Neighbors (KNN), Logistic Regression, Random Forest, and Support Vector Machine (SVM).
  • Quantum Models: Variational Quantum Classifier (VQC) with feature maps and ansatz circuits.

Features

• Feature engineering and selection using correlation thresholds.
• Class balancing using undersampling and SMOTE.
• Hyperparameter tuning with GridSearchCV.
• Quantum circuits for feature mapping and ansatz construction.
• Visualization of optimization progress during training.

Requirements

To set up and run the project, install the required dependencies listed in requirements.txt:

pip install -r requirements.txt

Project Structure

• Notebook: The .ipynb file contains the step-by-step analysis, modeling, and quantum circuit implementation.
• requirements.txt: Lists all dependencies for running the notebook.
• data/data.csv: The dataset used for analysis (replace with your data if necessary).

Key Dependencies

• Data Analysis & Visualization:
  • pandas, matplotlib, seaborn
• Classical Machine Learning:
  • scikit-learn, imblearn
• Quantum Machine Learning:
  • qiskit, qiskit-machine-learning
• Optimization:
  • scipy

Usage

1. Clone the Repository:

   git clone <repository_url>
   cd <repository_directory>

2. Install Dependencies:

   pip install -r requirements.txt

3. Run the Notebook: Open the .ipynb file in your preferred environment (e.g., Jupyter Notebook, VS Code) and execute the cells.

4. Explore Results: The notebook includes training results for classical models and quantum classifiers.

Results

• Classical Models:
  • Achieved F1-scores up to 0.83 on test data.
• Quantum Models:
  • Variational Quantum Classifier achieved comparable results with F1-scores around 0.81-0.83, demonstrating the feasibility of quantum machine learning for this task.

Future Improvements

• Explore additional quantum feature maps and ansatz configurations.
• Experiment with larger datasets and multi-class problems.
• Optimize quantum circuits for execution on real quantum hardware.

Author

• [Siddhant Gupta] - Quantum Machine Learning Researcher

---