Design of PSO-Based Hybrid Quantum Support Vector Machines for Breast Cancer Diagnosis

This repository presents a research project integrating Quantum Machine Learning (QML) with Meta-heuristic Optimization to improve diagnostic accuracy on the Wisconsin Breast Cancer dataset. The proposed PSO-QSVM architecture achieves 95.61% test accuracy, outperforming both Classical SVM and standard QSVM baselines.

---

💡 Motivation

Breast cancer diagnosis benefits from algorithms that can capture subtle, high-dimensional patterns in medical data. Classical SVMs hit a performance ceiling on this dataset around ~94% accuracy. We asked two questions:

1. Can quantum kernels capture patterns that classical kernels miss?
2. Can we automate the tedious hyperparameter tuning process?

The answer was a measurable +1.75% accuracy improvement (93.86% → 95.61%) — a difference that matters in clinical decision support, where every misclassification has a human cost.

---

🚀 Project Overview & Methodology

The core idea is to leverage the high-dimensional feature spaces of quantum mechanics to identify complex patterns in medical data that classical algorithms might miss.

• Quantum Data Encoding: Classical medical features are transformed into quantum states using Amplitude Encoding, allowing for efficient representation in Hilbert space.
• Hybrid Classification (QSVM): Unlike classical SVM, our model utilizes Quantum Kernels calculated on quantum circuits to measure data similarity with higher precision.
• Automated Optimization (PSO): To eliminate the inefficiency of manual tuning, the critical C-parameter is optimized using Particle Swarm Optimization (PSO), ensuring the model reaches its peak performance automatically.

---

📊 Comparative Performance Analysis

The proposed PSO-QSVM model was benchmarked against classical SVM and standard QSVM models. The results demonstrate a clear advantage:

Model Architecture	Test Accuracy
Classical SVM	93.86%
QSVM	94.74%
PSO-QSVM (Proposed)	95.61%

---

📋 Technical Specifications

• Dataset: Wisconsin Breast Cancer (Diagnostic) Dataset — 569 samples, 30 features
• Preprocessing: Min-Max Scaling and L² Normalization to satisfy quantum state requirements
• Stack: Python, Qiskit, PennyLane, scikit-learn, NumPy, Pandas
• Optimization: Particle Swarm Optimization for SVM C-parameter tuning

---

⚙️ Installation & Usage

# Clone the repository
git clone https://github.com/SidarYilmaz/Quantum_SVM_for_Breast_Cancer_Diagnosis.git
cd Quantum_SVM_for_Breast_Cancer_Diagnosis

# Create a virtual environment (recommended)
python -m venv venv
source venv/bin/activate    # On Windows: venv\Scripts\activate

# Install dependencies
pip install -r requirements.txt

# Launch Jupyter and run the notebooks
jupyter notebook notebooks/

---

📁 Repository Structure

.
├── notebooks/ # Jupyter notebooks (main experiments)
├── src/ # Reusable Python modules
├── experiments/results/ # Output metrics, plots, logs
├── requirements.txt # Python dependencies
└── README.md

---

👥 Research Team

This project was developed at Karadeniz Technical University:

• Prof. Dr. Tuğrul ÇAVDAR — Supervisor
• İrem Buse ÖZKÖSE — Researcher
• Sidar YILMAZ — Researcher

---

📚 Citation

If you use this work in your research, please cite:

@misc{yilmaz2025psoqsvm,
  title  = {Design of PSO-Based Hybrid Quantum Support Vector Machines for Breast Cancer Diagnosis},
  author = {Yılmaz, Sidar and Özköse, İrem Buse and Çavdar, Tuğrul},
  year   = {2025},
  institution = {Karadeniz Technical University},
  url    = {https://github.com/SidarYilmaz/Quantum_SVM_for_Breast_Cancer_Diagnosis}
}

---

📄 License

This project is licensed under the MIT License — see the LICENSE file for details.