Decision Boundary Plot Calculator

Decision Boundary Results

Load data, select a classifier, then calculate the boundary.

Train accuracy

—

Test accuracy

—

Macro F1

—

Log loss

—

Training time

—

Generalisation gap

—

Model summary

No model has been trained.

Confusion matrix

The matrix appears after calculation.

Prediction details

Prediction rows appear after calculation.

1. Dataset Input and Generation

Data rows: X1, X2, class label

Use commas, tabs, or semicolons. A header row is optional.

Synthetic dataset

Samples

Classes

Clusters per class

Class separation

Noise

Minority share

Random seed

2. Preprocessing and Validation

Feature scaling

Missing values

Test percentage

Decimals

Cross-validation

Shuffle before splitting

Use stratified split

Remove extreme outliers

Outlier z-threshold

3. Classifier and Parameters

Classification algorithm

Learning rate

Iterations / epochs

Regularization

Regularization strength

K neighbours

Distance metric

Minkowski power

Neighbour weighting

Tree depth

Minimum leaf size

Split criterion

Forest trees

SVM C

Gamma

Polynomial degree

Binary threshold

Hidden neurons

Batch size

Activation

Use intercept

4. Boundary and Plot Controls

Grid resolution

Axis padding

Region opacity 0.40

Point size 9

Point opacity 0.90

Chart theme

Plot title

X-axis label

Y-axis label

Show grid

Show legend

Equal axis scale

Confidence contours

Misclassified points

Train/test markers

Support-like points

5. Interactive Prediction and Comparison

Custom X1

Custom X2

Class for new training point

Click the plot or enter coordinates to test a prediction.

Models to compare

Hold Ctrl or Command to change multiple selections.

Model comparison results appear here.

Formula Used

Linear decision function: f(x) = w₁x₁ + w₂x₂ + b

Logistic probability: P(y = 1 | x) = 1 / (1 + e^-f(x))

Euclidean distance: d(x, z) = √[(x₁ − z₁)² + (x₂ − z₂)²]

SVM decision: class(x) = sign(w · x + b)

Accuracy: correct predictions / total predictions

F1 score: 2 × precision × recall / (precision + recall)

Generalisation gap: training accuracy − testing accuracy

How to Use

Paste data or generate a synthetic classification dataset.
Choose preprocessing, splitting, and validation settings.
Select a classifier and adjust its main parameters.
Customize the decision regions and point appearance.
Press Calculate and Plot to train the model.
Review metrics, predictions, warnings, and the confusion matrix.
Click the chart to test new coordinates interactively.
Export the chart, predictions, dataset, or PDF report.

Example Data

X1	X2	Class	Expected region
1.0	1.2	Class A	Lower-left region
2.5	1.6	Class A	Lower-middle region
4.4	4.2	Class B	Upper-middle region
6.1	4.9	Class B	Upper-right region

Frequently Asked Questions

What is a decision boundary?

It is the dividing line or region separating predicted classes.

Why does scaling change the boundary?

Distance-based and gradient-based models react strongly to feature scale.

Can the calculator handle multiclass data?

Yes. It supports multiple class regions and per-class metrics.

What causes an irregular boundary?

Noise, model complexity, limited samples, and nonlinear relationships can contribute.

How can I identify overfitting?

Look for high training accuracy and noticeably lower testing accuracy.

Why might KNN look jagged?

Small neighbourhood sizes create highly local and flexible regions.

What does confidence shading show?

It estimates how certain the selected model is across feature space.

Can I upload labels as words?

Yes. Numeric or text labels are accepted and encoded automatically.

Does the calculator replace production libraries?

No. It is an educational browser implementation for exploration and comparison.