Reading a Confusion Matrix
Explore how to read and interpret confusion matrices in classification evaluation. Understand the distinctions between true positives, false positives, true negatives, and false negatives, and learn how decision thresholds affect model outcomes. This lesson helps you translate classification results into meaningful business insights.
A fraud classifier makes one decision for each credit-card transaction. It either flags the transaction as fraud or allows it to proceed. A model that gets 19 of 20 transactions right may still create the wrong outcome if its one mistake is a stolen-card purchase that it approves. Another model with the same accuracy may instead block one legitimate purchase and inconvenience a customer.
Accuracy combines every correct and incorrect decision into one number. Fraud operations need the error types separated because missed fraud and unnecessary blocks have different costs. Evaluation begins by counting what happened to every transaction.
The four possible outcomes
For a fraud classifier, 1 means fraud and 0 means legitimate. Comparing the actual label with the model’s predicted label produces four outcomes.
A true positive (TP) occurs when a fraudulent transaction is flagged as fraud. The model caught fraud.
A false positive (FP) occurs when a legitimate transaction is flagged as fraud. The model ...