Support Vector Machine (SVM) Classification Project

Overview

This project implements Support Vector Machine (SVM) classifiers with both Linear and RBF kernels. It includes preprocessing, model training, hyperparameter tuning, evaluation, and optional 2D decision boundary visualization.

If no dataset is provided, it defaults to the Breast Cancer Wisconsin dataset from scikit-learn.

Features

• Load dataset from CSV (--csv) and specify target (--target), or use the default dataset.
• Preprocessing: handle missing values, drop obvious ID/unnamed columns, and encode non-numeric targets.
• Feature scaling with StandardScaler.
• Train Linear SVM and RBF SVM.
• Hyperparameter tuning for RBF SVM using GridSearchCV over C and gamma.
• Cross-validated metrics, confusion matrices, and classification reports saved to disk.
• Decision boundary plots when exactly two features are used.

Folder Structure

svm_classification_project/
│
├── data/                 
├── outputs/              
├── src/                  
│   └── svm_model.py
├── README.md
└── requirements.txt

Installation

python -m venv venv
# Windows
venv\Scripts\activate
# macOS/Linux
source venv/bin/activate

pip install -r requirements.txt

Usage

1) Default dataset (Breast Cancer)

python src/svm_model.py

2) Custom dataset

python src/svm_model.py --csv data/your_dataset.csv --target target_column_name

3) Custom dataset with exactly two features (enables decision boundaries)

python src/svm_model.py --csv data/your_dataset.csv --target target_column_name --features feature1 feature2

Outputs

• outputs/metrics.json – accuracy and best parameters
• outputs/confusion_matrix_linear.png
• outputs/confusion_matrix_rbf.png
• outputs/decision_boundary_linear.png (if 2D)
• outputs/decision_boundary_rbf.png (if 2D)
• outputs/accuracy_bar.png

Notes

• For categorical targets (e.g., "M"/"B"), the script automatically encodes them.
• Missing numeric values are imputed with the column mean.
• If you do not pass --target with a CSV, the script will prompt you to specify it.