“Your model says ‘SPAM’. But was it right? Your model says ‘NOT SPAM’. Was it correct? You need a scorecard. Not just one number. A full report of where your model wins and where it fails. That report is the Confusion Matrix.”
What Is a Confusion Matrix? #
A 2×2 table that shows exactly what your model got right and wrong.
The Table:
| Predicted: YES | Predicted: NO | |
|---|---|---|
| Actual: YES | TP (Hit) | FN (Miss) |
| Actual: NO | FP (False Alarm) | TN (Correct Reject) |
Four numbers. That is it.
The Four Numbers (Learn These) #
| Name | Plain English | Another Name |
|---|---|---|
| TP (True Positive) | Said YES, was YES | Hit |
| TN (True Negative) | Said NO, was NO | Correct Reject |
| FP (False Positive) | Said YES, was NO | Type I Error, False Alarm |
| FN (False Negative) | Said NO, was YES | Type II Error, Miss |
Memory Trick:
- “True” = Model was correct
- “False” = Model was wrong
- “Positive” = Model said YES
- “Negative” = Model said NO
So “False Positive” = Model was wrong when it said YES.
Real Example: Cancer Test #
1000 patients tested. 100 actually have cancer. 900 are healthy.
Model Results:
| Predicted Cancer | Predicted Healthy | |
|---|---|---|
| Actually Cancer | 80 (TP) | 20 (FN) |
| Actually Healthy | 30 (FP) | 870 (TN) |
Read the matrix:
- TP = 80 → Correctly caught 80 cancer patients
- FN = 20 → Missed 20 cancer patients (dangerous!)
- FP = 30 → Told 30 healthy people they have cancer (stressful)
- TN = 870 → Correctly told 870 healthy people they are fine
One Real Example #
Cancer test. 100 sick people. 900 healthy.
Model results:
- Found 80 sick correctly (TP)
- Missed 20 sick (FN)
- Told 30 healthy they have cancer (FP)
- Told 870 healthy they are fine (TN)
That is it. The whole matrix.
Why Bother? #
One number like Accuracy hides the truth.
Accuracy says 95% (870+80 = 950 correct out of 1000).
But you missed 20 sick people. That could kill someone.
Confusion matrix shows you the problem.
Every Metric Comes From Here #
| You Want | Look At |
|---|---|
| Correct predictions | TP + TN |
| False alarms | FP |
| Missed catches | FN |
| Trust when model says YES | TP / (TP + FP) → Precision |
| How many caught | TP / (TP + FN) → Recall |
One Line of Code #
from sklearn.metrics import confusion_matrix print(confusion_matrix(y_true, y_pred))
Key Takeaways (3 Lines) #
- Confusion matrix shows exactly what your model got right and wrong.
- Two errors matter: False alarms (FP) and Missed catches (FN).
- Always check the matrix before trusting accuracy.