Confusion Matrix Calculator - Accuracy, Precision, Recall, and F1

A confusion matrix calculator turns the four-cell summary of a binary classification result into the standard evaluation metrics used in machine learning, statistics, and academic work. Type the true positives, false negatives, false positives, and true negatives, and the tool returns accuracy, precision, recall, F1, specificity, false positive and false negative rates, prevalence, and the Matthews correlation coefficient in one step.

• • Grading a homework classifier: Compare a student logistic regression against a target confusion matrix from a textbook without retyping the formulas.
• • Evaluating a medical screening tool: Check sensitivity, specificity, and the Matthews correlation coefficient where false negatives carry real cost.
• • Tuning a spam or fraud filter: See how a new threshold changes precision, recall, and F1 when the team debates flagging more messages.
• • Comparing two model versions: Drop the confusion matrix counts from each candidate model side by side and compare every metric at once.

Most reports print a 2-by-2 matrix of TP, FN, FP, and TN, but the derived metrics are scattered across dashboards. A single page that takes the four counts and turns them into the standard ratios makes reading a model quicker.

The top row (TP and FN) covers actual positive cases, and the bottom row (FP and TN) covers actual negative cases. The left column (TP and FP) holds the model's positive predictions, and the right column (FN and TN) holds the negative ones.

For a broader view of descriptive statistics on the same dataset, the Statistics Calculator accepts a raw list of values and reports mean, median, variance, and standard deviation alongside it.

The calculator reads the four cell counts, sums them into a total, and divides the right combinations to produce each ratio. Denominators of zero are treated as undefined so the page does not silently misreport metrics that have no meaning in your matrix.

• TP: True positives: actual positives predicted as positive.
• FN: False negatives: actual positives predicted as negative.
• FP: False positives: actual negatives predicted as positive.
• TN: True negatives: actual negatives predicted as negative.
• N: Total samples: TP + FN + FP + TN.

Precision answers 'when the model predicts positive, how often is it right?' and recall answers 'of all actual positives, how many did the model catch?' They are almost always a trade-off, which is why the F1 score, the harmonic mean of the two, is so widely used.

The Matthews correlation coefficient (MCC) is harder to compute by hand because it uses all four cells and a product of four sums in the denominator. According to Wikipedia, accuracy is the proportion of correct predictions, precision is TP divided by (TP plus FP), and recall is TP divided by (TP plus FN).

According to Wikipedia: Confusion matrix, accuracy is the proportion of correct predictions, precision is TP divided by (TP plus FP), and recall is TP divided by (TP plus FN).

If you want to manipulate the four cells as a real 2-by-2 array, the Matrix Calculator lets you add, multiply, and invert matrices built from the same numbers.

These four concepts are the vocabulary that lets you talk about a confusion matrix calculator result out loud. Skim the names first, then read the explanations once you have a count to look at.

Once you can name these four ideas, the rest of the confusion matrix vocabulary - false discovery rate, false omission rate, balanced accuracy - falls into place as a small rearrangement of the same four cells.

Many textbooks pair a confusion matrix with a hypothesis test. The chi-square test uses the same observed-versus-expected structure, so the four cells you type here often feed straight into a chi-square analysis.

When you want to know whether the pattern in the four cells is statistically significant, the Chi-Square Calculator turns the same observed counts into a chi-square statistic and a p-value.

Five quick steps cover the typical workflow: pull the four counts out of your report, type them in, read the result panel, scan for undefined metrics, then screenshot your write-up.

1. 1 Locate the four counts in your report: Open the classification report or library output that shows TP, FN, FP, and TN. If you only have class-level precision and recall, use the formulas on this page to back out the counts.
2. 2 Enter the four counts: Type the integer value of each cell into the matching input. Use whole numbers - the calculator rounds down any decimal so the metrics always match a real count.
3. 3 Read the result panel: Watch the result panel update as you type. Accuracy, precision, recall, F1, specificity, false positive and negative rates, prevalence, and the Matthews correlation coefficient all update from one recalculation.
4. 4 Spot divide-by-zero issues: If a metric shows zero where you expected a real value, check the denominator. Precision needs at least one predicted positive (TP + FP greater than zero) and recall needs at least one actual positive (TP + FN greater than zero).
5. 5 Save and compare to a baseline: Copy the result table into your lab notebook, slide deck, or pull request, and run the same matrix through a baseline model. The side-by-side comparison makes it much easier to argue a new threshold is moving the metrics you care about.

Example workflow: read off TP = 50, FN = 10, FP = 5, TN = 100 from a confusion matrix image, type them in, and the panel reports accuracy 0.91, precision 0.91, recall 0.83, F1 0.87, MCC 0.81 - the same numbers you would get by hand.

If you are comparing two confusion matrices from two different models, the T-Test Calculator tells you whether the difference in their accuracies is large enough to be statistically meaningful.

These are the concrete benefits students, instructors, and machine learning practitioners report from a single-page tool.

• • Cuts the metric confusion: Every standard classification metric appears in the same place, so you stop mixing up precision with recall, sensitivity with specificity, or F1 with accuracy.
• • Makes imbalanced classes honest: By showing F1 and MCC alongside accuracy, the result panel reveals when a high accuracy number hides a useless model, a common failure on imbalanced datasets.
• • Saves a stats homework step: Enter four cells once and copy the row of metrics straight into the answer box instead of typing formulas into a spreadsheet.
• • Speeds up model reviews: When a teammate pastes a confusion matrix in a pull request, the reviewer can punch in the four numbers and confirm the metrics within seconds.
• • Reinforces the matrix layout: Typing the counts by hand forces you to remember which cell is TP and which is FN, the most common source of bug in student code.

The biggest pay-off is the first time you build a model that scores 99 percent accuracy on a credit-card-fraud dataset and look at the F1 to see the model is predicting 'not fraud' for almost everything. The result panel makes that surprise visible in seconds.

For instructors, putting the page on a class site lets students check homework without giving them the answer key.

If the four cells come from a randomized experiment with a known positive rate, the Binomial Distribution Calculator helps you decide whether the observed false negative count is unusual under the null hypothesis.

Three factors change the numbers you see in the result panel, and two limitations remind you which metrics are safe to quote and which need extra context.

• • This tool is built for binary classification. Multi-class problems have to be collapsed into one-versus-rest counts before the four cells make sense.
• • Counts are entered as integers and rounded down. If your pipeline reports floating point counts, normalize to the nearest whole number first.

If you need a single number to summarize a binary classifier on an imbalanced dataset, prefer the Matthews correlation coefficient over accuracy. It is symmetric across the two classes, so swapping the positive label only changes the sign.

The matrix is only as honest as the labels you started with. If the ground-truth labels are noisy, common in medical imaging or user-tagged datasets, the matrix inherits that noise.

As published by Google Machine Learning Crash Course, accuracy can be misleading for imbalanced classes and a high recall or high F1 score is often more informative when positive cases are rare.