Interactive RBF Kernel SVM Sandbox

Red Prediction Space

Blue Prediction Space

Decision Boundary

Confusion Matrix (Training Set Evaluation)

Predicted Red (+)

Predicted Blue (-)

Total

Actual Red

0 0.0% True Red (TP)

0 0.0% False Blue (FN)

Actual Blue

0 0.0% False Red (FP)

0 0.0% True Blue (TN)

Total

1. Select Class to Plot

Class +1 (Red) Class -1 (Blue)

2. Kernel Sensitivity (Gamma: γ)

γ Coefficient15.0

Higher values create highly localized decision pockets (potential overfitting). Lower values create broader, sweeping boundaries.

3. Data Dispersion (Noise Level)

Variance Amplitude15

Alters the radial spread and interlocking variance of the generated sample rings.

4. Select Target Focus Class for Metrics Calculations

Red Class (+) Blue Class (-)

Derived Evaluation Metrics Calculations

Accuracy

0.0%

Precision

0.0%

Sensitivity (Recall)

0.0%

Specificity

0.0%

Model Status: Optimization Active
Plot coordinates to watch the RBF kernel dynamically snake around the clusters.

🔬 Classroom Experiment: Push Gamma up to 50 and Noise up to 35. Notice how the boundary contracts into tight isolations around individual nodes. Observe how this complex behavior shifts Accuracy closer to 100% on this training set, visually demonstrating training-set overfitting as localized radial pockets multiply!