NeuroNomixer logoNeuroNomixer
  • Home
  • Blog
  • Visual Guides
  • Authors
  • Contact
Sign InSign Up
NeuroNomixer logo
HomeBlogAuthorsContactPrivacy Policy

© 2026 NeuroNomixer — Built with Next.js & Tailwind CSS

Visual Guides/The Confusion Matrix Decoded
Machine Learning

The Confusion Matrix Decoded

Accuracy alone doesn't tell the whole story. Explore how the confusion matrix reveals what kinds of mistakes your classifier makes, and why that matters differently in each domain.

Exploration Progress1/3 scenarios · 0/5 threshold changes

Sign in to save your progress.

🩺Medical Diagnosis

False negatives (missed cancer) are more dangerous than false positives (unnecessary tests).

Confusion Matrix at threshold = 0.50

Predicted: Healthy
Predicted: Cancer
Actual: Healthy
127
TN
63.5%
33
FP
16.5%
Actual: Cancer
8
FN
4.0%
32
TP
16.0%
Accuracy
79.5%
(TP+TN)/Total
Precision
49.2%
TP/(TP+FP)
Recall (TPR)
80.0%
TP/(TP+FN)
Specificity
79.4%
TN/(TN+FP)
F1 Score
61.0%
2·P·R/(P+R)
FPR
20.6%
FP/(FP+TN)

Scenario

Classification Threshold

0.50

Samples with score ≥ threshold are predicted positive. Lower threshold → more positives predicted.

0.05 (predict +)0.95 (predict −)

Threshold Tradeoff

↓ threshold
↑ Recall (catch more positives)
↑ FP rate (more false alarms)
↑ threshold
↑ Precision (fewer false alarms)
↓ Recall (miss more positives)

Key Insight

In medical screening, a missed cancer (FN) can be fatal, so you lower the threshold to maximize recall, accepting more false alarms. In spam filtering, you'd rather let spam through than block a job offer.

← All GuidesNext Guide →