SVM Comparison

Why Try SVM?

SVMs are powerful classifiers capable of separating data with non-linear boundaries. While slower to train, they can excel in datasets where class clusters are tightly shaped and margins matter.

Kernel Selection and Scaling

A radial basis function (RBF) kernel was selected for its ability to model complex curvature. Features were scaled using the same standardization strategy as logistic regression to ensure fair comparison.

from sklearn.svm import SVC

svm = SVC(kernel="rbf", C=1.0, gamma="scale")
svm.fit(X_train, y_train)

The model was cross-validated using the same stratified training and test splits used throughout the project.

Performance and Confusion Matrix

On the held-out test set, the SVM achieved comparable results to logistic regression, with minor differences in sensitivity to borderline cases.

              precision    recall  f1-score   support

       Benign       0.97      0.99      0.98       72
    Malignant       0.97      0.93      0.95       42

    accuracy                           0.97      114
   macro avg       0.97      0.96      0.96      114
weighted avg       0.97      0.97      0.97      114

These results indicate that SVM slightly improved the false positive rate for malignant predictions, which could be beneficial in sensitive diagnoses.

Model Comparison

Both classifiers showed high accuracy, but logistic regression had the advantage of transparency and faster training. SVM's nonlinear boundary gave it a slight edge in recall.

Key Takeaways

• SVM delivered strong results on the same input features as logistic regression.
• Radial kernels captured subtle patterns without needing feature transformation.
• Interpretability was reduced compared to the linear model.
• Both models demonstrated the diagnostic value of structured FNA data.