Results will appear here
Load the example or enter classification data, then select Calculate.
Model Summary
Confusion Matrix Heatmap
diagonal cells are correct predictions; off-diagonal cells are errors.
Select a heatmap cell to inspect its value.
Matrix Values
Per-Class Metrics
Binary Classification Metrics
Misclassification Analysis
Observation-Level Data
Model Comparison
Automatic Interpretation
Calculator Inputs
Use commas, spaces, tabs, semicolons, or new lines.
The number of predicted labels must match actual labels.
Enter probability data to evaluate threshold performance.
Formula Used
Accuracy
(TP + TN) ÷ (TP + TN + FP + FN)
(TP + TN) ÷ (TP + TN + FP + FN)
Precision
TP ÷ (TP + FP)
TP ÷ (TP + FP)
Recall
TP ÷ (TP + FN)
TP ÷ (TP + FN)
Specificity
TN ÷ (TN + FP)
TN ÷ (TN + FP)
F1 Score
2 × Precision × Recall ÷ (Precision + Recall)
2 × Precision × Recall ÷ (Precision + Recall)
Balanced Accuracy
Average recall across all classes
Average recall across all classes
Cohen’s Kappa
(Observed accuracy − Expected accuracy) ÷ (1 − Expected accuracy)
(Observed accuracy − Expected accuracy) ÷ (1 − Expected accuracy)
Multiclass MCC
Correlation-style score derived from the complete confusion matrix
Correlation-style score derived from the complete confusion matrix
How to Use
- Enter actual and predicted labels, upload CSV data, or build a manual matrix.
- Select normalization, palette, scaling, decimal precision, and positive class options.
- Choose Calculate Heatmap to generate the matrix and performance metrics.
- Inspect cells, filter observations, compare snapshots, and export your results.
Example Data
| Observation | Actual | Predicted |
|---|---|---|
| 1 | Cat | Cat |
| 2 | Dog | Cat |
| 3 | Bird | Bird |
| 4 | Cat | Dog |
| 5 | Dog | Dog |
Frequently Asked Questions
What does the diagonal represent?
Diagonal cells contain correct predictions for each class.
What are off-diagonal cells?
They show one class being incorrectly predicted as another.
When should I normalize by rows?
Row normalization compares prediction behavior within each actual class.
When should I normalize by columns?
Column normalization examines the composition of each predicted class.
Why can a metric be undefined?
Its denominator may be zero because a class lacks observations or predictions.
Does high accuracy prove model quality?
No. Class imbalance can make accuracy look stronger than practical performance.
Which average should I use?
Macro treats classes equally, weighted uses support, and micro pools all decisions.
Can I compare models?
Yes. Save snapshots and review differences in the comparison table.
Is uploaded data stored?
No. Calculations run locally in your browser.
Can I use text class labels?
Yes. Numeric, text, and mixed labels are supported.