SVM from Scratch: Gradient Descent vs Subgradient Descent

This project implements a Support Vector Machine (SVM) classifier from scratch using Hinge Loss on the Banknote Authentication dataset. It compares the performance of Gradient Descent and Subgradient Descent optimization techniques.

---

📁 Project Structure

.
├── data_banknote_authentication.csv    # Dataset
├── main.py                             # Entry point for training and evaluation
├── requirements.txt                    # Required Python packages
└── README.md                           # Project documentation

---

📊 Features

• SVM implementation from scratch (no scikit-learn)
• Hinge Loss function (non-differentiable)
• Gradient Descent and Subgradient Descent optimizers
• Tracks and plots:
  • Loss per epoch
  • Training and validation accuracy
  • Confusion matrix
• Evaluation on convergence, accuracy, and generalization

---

📈 Visualizations

• Loss Curve (per epoch)
• Accuracy Curve (training & validation)
• Confusion Matrix

---

🧪 Evaluation Metrics

• Training Accuracy
• Validation Accuracy
• Convergence Speed
• Loss Stability
• Generalization to unseen data

---

📦 Installation

git https://github.com/MuhammedMahmoud0/svm-subgradient-vs-gradient.git
cd svm-subgradiant-vs-gradiant
pip install -r requirements.txt