How do I read a confusion matrix?

Rows are actual classes, columns are predictions. Diagonal shows correct predictions. Off-diagonal shows errors. Read: "Actual X was predicted as Y" by finding row X, column Y.

What does an ideal confusion matrix look like?

Diagonal elements (correct predictions) are high, off-diagonal (errors) are low or zero. In reality, some errors always exist - the goal is minimising the errors that matter most for your use case.

How do I use confusion matrix for multi-class problems?

Same concept, larger matrix. N classes → N×N matrix. Examine which classes get confused with each other. May reveal that certain similar classes need more distinguishing features.

Should I normalise the confusion matrix?

Often yes - normalise by row (actual) to see recall per class, or by column (predicted) to see precision per class. Raw counts can be misleading with imbalanced classes.

Book Free Assessment Calculator

Confusion Matrix

A table showing predicted vs actual classifications, revealing true positives, false positives, true negatives, and false negatives. Essential for understanding model error patterns.

In-Depth Explanation

A confusion matrix is a table that describes classification model performance in detail. It shows how predictions map to actual outcomes, revealing exactly what types of errors the model makes.

Binary confusion matrix: Predicted Pos Neg Actual Pos TP FN Neg FP TN

Key components:

True Positive (TP): Correctly predicted positive
False Positive (FP): Incorrectly predicted positive
True Negative (TN): Correctly predicted negative
False Negative (FN): Incorrectly predicted positive

Derived metrics:

Accuracy = (TP + TN) / Total
Precision = TP / (TP + FP)
Recall = TP / (TP + FN)
Specificity = TN / (TN + FP)

Multi-class extension:

Rows: actual classes
Columns: predicted classes
Diagonal: correct predictions

Business Context

Confusion matrices reveal WHERE the model fails. A single accuracy number hides whether errors are false positives or false negatives - which have very different business implications.

How Clever Ops Uses This

We always examine confusion matrices for Australian business ML projects to understand error patterns and tune models appropriately.

Example Use Case

"Confusion matrix reveals the churn model misclassifies high-value customers more often - informing targeted improvement efforts."

Frequently Asked Questions

Related Terms

Precision Recall Accuracy

Learn More

Model Selection and Evaluation: Choosing the Right AI Model for Your Use Case

Learn how to select the optimal AI model for your needs by comparing capabilities, costs, and performance. Includes evaluation frameworks, benchmarking strategies, and migration guidance.

Read article

Conflict of Interest Connector