SVM Classifier

---

🎯 Purpose

This QuickRef outlines how to use Support Vector Machine (SVM) classifiers. It covers decision boundaries, kernels, hyperparameters, and practical usage for binary and multiclass classification.

---

📦 1. When to Use

Condition	Use SVM?
Linear boundary works well	✅ Yes (linear SVM)
You need clear margin between classes	✅ Yes
Nonlinear boundary needed	✅ Yes (use kernels)
Huge dataset or sparse input	❌ Can be slow; use SGD or trees

---

✂️ 2. Core Logic

• Finds hyperplane that maximally separates classes
• Only support vectors (borderline points) influence boundary
• Uses kernel trick to map inputs to higher dimensions

---

🛠️ 3. Fitting in sklearn

from sklearn.svm import SVC
model = SVC(kernel='rbf', C=1.0, probability=True)
model.fit(X_train, y_train)

✔️ Use probability=True for predict_proba() (not on by default)

---

🔧 4. Key Hyperparameters

Param	Description
C	Regularization strength (low = wider margin, more misclassifications)
kernel	Mapping function: 'linear', 'rbf', 'poly', 'sigmoid'
gamma	Controls curvature of decision boundary (higher = more flexible)
probability	Enables predict_proba() (slower training)

---

🌐 5. Common Kernels

Kernel	Use When...
linear	Data is roughly linearly separable
rbf (default)	General-purpose nonlinear mapping
poly	Polynomial relationships exist
sigmoid	Rarely used (like a shallow neural net)

---

📊 6. Evaluation Tips

• Use accuracy, precision, recall, AUC
• Use GridSearchCV to tune C and gamma
• Normalize or standardize features before fitting

---

✅ Checklist

• [ ] Target is binary or multiclass
• [ ] Feature scaling applied
• [ ] Kernel chosen based on linearity needs
• [ ] C and gamma tuned via cross-validation
• [ ] If predict_proba needed, enable in SVC

---

💡 Tip

“Support Vector Machines don’t memorize your data — they focus only on what matters: the boundary.”