Confusion Matrix Calculator for Machine Learning

Evaluation Results

Enter data and calculate a confusion matrix.

Confusion Matrix

Display

Matrix Heatmap

Per-Class Metrics

Metric Comparison

Interpretation

No interpretation is available yet.

Validation and Warnings

No validation messages.

Formula Used

Formulas appear after calculation.

Step-by-Step Calculation

Calculation steps appear here.

Saved Model Comparison

Save calculated model snapshots to compare performance.

1. Input Method

Positive class name

Negative class name

True positives

True negatives

False positives

False negatives

Actual labels

Predicted labels

Delimiter

Comparison

Positive class for binary metrics

Ignore blank values

Example dataset

Actual binary labels

Positive-class probabilities

Positive label

Negative label

Threshold: 0.50

Optimise threshold for

Class names

Separate class names with commas.

CSV or TSV file

File delimiter

First row contains headers

Actual-label column

Predicted-label column

Probability column

Sample-weight column

2. Metric and Cost Options

F-beta value

Decimal precision

Metric format

Primary multiclass average

True-positive value or cost

True-negative value or cost

False-positive cost

False-negative cost

Calculate automatically

The active input tab controls the calculation source.

How to Use

Select binary counts, labels, probabilities, manual matrix, or file import.
Enter valid classification data and choose calculation options.
Press calculate to generate the matrix and metrics.
Review warnings, interpretation, formulas, and class-level results.
Copy, print, or export the completed evaluation report.

Example Data

Actual	Predicted	Meaning
Positive	Positive	True positive
Negative	Negative	True negative
Negative	Positive	False positive
Positive	Negative	False negative

Frequently Asked Questions

What is a confusion matrix?

It compares actual classes with predicted classes. Correct results appear on the main diagonal. Errors appear outside that diagonal.

Which axis represents actual classes?

This calculator displays actual classes as rows. Predicted classes appear as columns. Labels remain visible in every matrix.

What happens when a denominator is zero?

The affected metric displays N/A. This avoids misleading infinite or undefined values. A warning explains the issue clearly.

Why can accuracy be misleading?

Large majority classes can dominate overall accuracy. Minority performance may still remain poor. Review recall and balanced accuracy too.

What is macro averaging?

Macro averaging gives every class equal importance. It averages each class metric directly. Class frequency does not change weighting.

What is weighted averaging?

Weighted averaging uses class support as weights. Larger classes influence results more strongly. It reflects the observed class distribution.

What is micro averaging?

Micro averaging combines decisions across all classes. It totals contributions before calculating metrics. Larger classes naturally contribute more.

How is the best threshold selected?

The calculator evaluates candidate probability thresholds. It compares the chosen optimisation metric. The strongest valid threshold is selected.

Can I import spreadsheet data?

Yes, upload CSV or TSV files. Select actual and predicted columns. Optional probability and weight columns are supported.

Evaluation Results