Confusion Matrix Calculator
Calculate Accuracy, Precision, Recall, F1, Specificity, NPV, FPR, FNR and Matthews Correlation Coefficient from a binary classifier's TP/TN/FP/FN counts, with the formula and a quality rating shown for every metric.
Input
Correctly predicted positive cases.
Actual positives the model missed (predicted negative).
Actual negatives the model flagged as positive.
Correctly predicted negative cases.
Output
Your results will appear here.
| Predicted Positive | Predicted Negative | |
|---|---|---|
| No data yet | ||
| Metric | Formula | Value | Rating |
|---|---|---|---|
| No data yet | |||
More ways to use this tool
REST API
curl -X POST https://api.iotools.cloud/v1/tool/confusion-matrix-calculator \
-H "Authorization: Bearer YOUR_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"truePositive": "50",
"falseNegative": "10",
"falsePositive": "5",
"trueNegative": "100"
}'Swap in your own key from your account. The tool's fields are the body — no wrapper.
Ask an AI agent
Use the IOTools `confusion-matrix-calculator` tool (Confusion Matrix Calculator) on this input:
YOUR_INPUT_HEREPaste this at any agent connected to the IOTools MCP server, then add your input.
Embed widget
<iframe
src="https://iotools.cloud/embed/confusion-matrix-calculator/"
width="100%" height="520" frameborder="0" scrolling="no" loading="lazy"
title="Confusion Matrix Calculator — iotools.cloud"
sandbox="allow-scripts allow-forms allow-same-origin allow-downloads allow-popups allow-popups-to-escape-sandbox"
allow="clipboard-write"
style="width:100%;border:1px solid #e5e7eb;border-radius:12px;overflow:hidden"></iframe>
<script src="https://iotools.cloud/embed.js" async></script>Drop this into your own page — free, no key required, just a link back.
| Cost per API/MCP call | From 5 credits |
|---|---|
| Need more credits? | View pricing |
Also available with
Guides
What does this calculator do?
Enter the four counts from a binary classifier's confusion matrix — True Positive, False Negative, False Positive, and True Negative — and get every standard evaluation metric computed instantly: Accuracy, Precision, Recall, Specificity, F1 Score, Matthews Correlation Coefficient (MCC), Negative Predictive Value, False Positive Rate, False Negative Rate, and a Balanced Accuracy estimate. Each metric shows its formula and a quality rating (Excellent / Good / Fair / Poor), so you don't have to remember which formula is which or re-derive them by hand.
How to use it
- Enter True Positive (TP) — cases correctly predicted positive.
- Enter False Negative (FN) — actual positives the model missed.
- Enter False Positive (FP) — actual negatives the model flagged as positive.
- Enter True Negative (TN) — cases correctly predicted negative.
- Read the results: an overall quality gauge, the matrix laid out as a table, and every metric with its formula and rating.
Why use MCC instead of just Accuracy?
Accuracy can be misleading on imbalanced datasets — a classifier that always predicts "negative" on a dataset that's 99% negative scores 99% accuracy while catching zero positives. Matthews Correlation Coefficient uses all four confusion-matrix counts in one balanced measure, ranging from -1 (total disagreement) to +1 (perfect prediction), with 0 meaning no better than random guessing — which is why it's considered the more reliable single number for imbalanced classes.
What's the difference between Precision and Recall?
Precision answers "of everything I predicted positive, how much was actually positive?" (TP / (TP+FP)) — it penalizes false alarms. Recall (Sensitivity) answers "of everything that was actually positive, how much did I catch?" (TP / (TP+FN)) — it penalizes misses. The two trade off against each other, which is why F1 Score (their harmonic mean) is often reported as a single balance point between them.
Is the "AUC estimate" a real AUC?
Not exactly — a true ROC AUC needs classifier scores across many decision thresholds, which a single confusion matrix (one threshold) can't provide. The "AUC estimate" shown here is Balanced Accuracy — the average of Recall and Specificity — a common single-threshold approximation, not a substitute for computing a real ROC curve if you have per-sample scores.
Privacy
All calculations run locally in your browser — your counts are never sent to a server.
For evaluating whether a difference between two rates is statistically significant (e.g. comparing two models' conversion or error rates), see the A/B Test Significance Calculator. To plan how much data you'd need to detect a given effect size before collecting it, see the A/B Test Sample Size Calculator.