Scratch Coding Classes for Kids in Pilsen

Chapter 119 – Build an SVM Model with Scikit-Learn

1. End-to-End Pipeline

1. Load data (Breast Cancer: 569 samples, 30 features, binary labels)
2. Split 75% train / 25% test
3. Fit Linear, Poly, and RBF SVC models
4. Tune C (e.g. 0.1, 1, 10, 50)
5. Report accuracy, confusion matrix, precision, recall, F1
6. Scale features before distance-based kernels

2. Load and Split

from sklearn.datasets import load_breast_cancer
from sklearn.model_selection import train_test_split

cancer = load_breast_cancer()
X, y = cancer.data, cancer.target
X_train, X_test, y_train, y_test = train_test_split(
    X, y, test_size=0.25, random_state=42
)
# ~426 train, ~143 test

3. Kernel Comparison (Unscaled)

from sklearn.svm import SVC
from sklearn.metrics import accuracy_score, confusion_matrix

for name, model in {
    'Linear': SVC(kernel='linear', random_state=42),
    'RBF': SVC(kernel='rbf', random_state=42),
    'Poly': SVC(kernel='poly', random_state=42)
}.items():
    model.fit(X_train, y_train)
    pred = model.predict(X_test)
    print(name, accuracy_score(y_test, pred), confusion_matrix(y_test, pred))

Typical results: Linear ~95.1% (few errors), Poly ~91.6%, RBF ~63% without scaling — RBF suffers when feature scales differ. Linear often wins on this dataset (moderate complexity).

4. Tuning C on RBF

Trying C ∈ {0.1, 1, 10, 50}: low C stays around ~63%; C=10 and C=50 reach about ~63.6%. Still weak until features are scaled.

5. Production Tips

• Always scale (e.g. StandardScaler) before RBF/Poly.
• Use GridSearchCV over kernel and C with cross-validation.
• Report accuracy plus confusion matrix, precision, recall, and F1.