AboutBackgroundAnalysisModelImagesAPIResults
Model workshop · SVMTry the image workshop

Probe the model.
See what it sees.

Switch kernels on the right, then drag the sliders below. The decision boundary, your prediction, and the surrounding region update live. Pick any two features as visualisation axes — the remaining eight are held at the slider values.

Features

10 inputs
Radius (mean)X axis
14.8 mm
6.0RFE rank 1 · importance 92%30.0
Texture (mean)
19.8
9.0RFE rank 6 · importance 41%40.0
Perimeter (mean)
96.8 px
40.0RFE rank 2 · importance 90%200.0
Area (mean)
721 px²
140RFE rank 3 · importance 88%2500
Smoothness (mean)
0.098
0.050RFE rank 9 · importance 23%0.170
Compactness (mean)
0.113
0.020RFE rank 7 · importance 38%0.360
Concavity (mean)
0.103
0.000RFE rank 4 · importance 81%0.430
Concave points (mean)Y axis
0.057
0.000RFE rank 5 · importance 78%0.210
Symmetry (mean)
0.183
0.100RFE rank 8 · importance 27%0.310
Fractal dim. (mean)
0.0628
0.0450RFE rank 10 · importance 9%0.1000

Decision boundary

Live feature space for the selected kernel. Hover any point for details.

Prediction
Benign

Based on RBF kernel, the model classifies this profile as benign with 68.4% confidence.

benignP(malignant) = 31.6%malignant
Signed decision
-0.2274
Distance to boundary
0.2274
Latency (live)
—
Top features pulling this prediction
Concave pts.
0.057 · → benign
Concavity
0.103 · → benign
Perimeter
96.8 · → benign

Model-specific stats

SVM — RBF kernel · 78 support vectors
98.3%
Accuracy
held-out test split
99.8%
ROC-AUC
area under ROC curve
97.6%
F1 Score
harmonic precision / recall
1.2 ms
Prediction latency
single-sample inference
97.6%
Precision
TP / (TP+FP)
97.6%
Recall
TP / (TP+FN)
78
Support vectors
trained in 23 ms
0.227
Distance to boundary
malignant prob. 31.6%

Confusion matrix

n = 114 · 98.2% acc
pred · malignant
pred · benign
actual · malignant
41tp
1fn
actual · benign
1fp
71tn
true positives 41false negatives 1false positives 1true negatives 71

Learning curve

accuracy vs. epoch
traintest

Bias / variance diagnostic

error vs. complexity (C)
bias (train err.)variance (cv err.)

Hyperparameters

K(xᵢ, xⱼ) = exp(−γ ||xᵢ − xⱼ||²)
C
10
γ (gamma)
0.01
kernel
rbf

Explainability

RAG retrieval over reference clinical documents

Why this prediction?

local summary
  • Using the svm — rbf kernel, your input falls in the benign region of feature space with 68.4% confidence (signed decision -0.227).
  • This input lies near the boundary — small movements in concavity or radius can flip the prediction. We recommend reviewing the bias-variance panel.
  • Your input sits at the neutral midpoint of every feature. Try adjusting radius, perimeter, area, or concave points to see how the prediction responds.
POST /api/explain
Predictions are advisory and require clinical validation before acting on them.

Retrieved reference passages

local catalog
01_wolberg_mangasarian_1990
Wolberg & Mangasarian (1990): Multisurface method for breast tumor diagnosis
Across 569 fine-needle aspirates the mean radius, perimeter, and concave-points features were the most discriminative — together they reach > 97% cross-validated accuracy with a linear classifier.
radius_meanperimeter_meanconcave_points_mean
04_guyon_rfe_2002
Guyon et al. (2002): Recursive Feature Elimination for SVMs
After RFE only six 'mean' features survive the 0.05 significance cut; smoothness and fractal dimension consistently drop out as redundant.
radius_meanconcavity_meansmoothness_meanfractal_dimension_mean
03_cortes_vapnik_svm
Cortes & Vapnik (1995): Support-Vector Networks
Inputs whose |f(x)| sits below ~0.4 are borderline — small perturbations of the most important features can flip the prediction.
decision_boundaryrbflinear

Input snapshot

JSON · ready to send to /api/predict
{
  "model": "rbf",
  "kernel": "rbf",
  "features": {
    "radius_mean": 14.85,
    "texture_mean": 19.75,
    "perimeter_mean": 96.75,
    "area_mean": 720.6,
    "smoothness_mean": 0.09770000000000001,
    "compactness_mean": 0.11265,
    "concavity_mean": 0.10305,
    "concave_points_mean": 0.056850000000000005,
    "symmetry_mean": 0.1835,
    "fractal_dimension_mean": 0.0628
  },
  "prediction": {
    "label": "benign",
    "probability_malignant": 0.3158,
    "confidence": 0.6842,
    "decision": -0.2274
  }
}