Inbox sorting arena
Drag the email card into Spam or Legitimate, or use the buttons and keyboard shortcuts.
Current email
Classifier laboratory
Selected features
Rule builder
Build weighted rules, then compare performance before and after applying them.
Challenge objectives
Achievements
Performance overview
Confusion matrix
Threshold trade-off
Feature importance
Probability distribution
Player versus model
Classification history
| # | Subject | Actual | Player | Model | Probability | Result | Actions |
|---|
Dataset tools
Custom fictional message
How the classifier learns
Binary classification assigns each fictional email to spam or legitimate. Training labels teach the model which patterns appear in each class. Testing data checks whether those patterns generalise.
Precision measures how often spam predictions are correct. Recall measures how much actual spam is detected. F1 balances both values when one score is needed.
Thresholds change the trade-off between missed spam and blocked legitimate messages. Lower thresholds usually increase recall. Higher thresholds often reduce false positives.
Class imbalance can make accuracy misleading. A model can seem accurate by predicting the dominant class. Balanced accuracy and confusion matrices expose that problem.
Safe inbox habits
- Verify unexpected requests through a separate trusted channel.
- Do not open suspicious links or unknown attachments.
- Check sender domains carefully for subtle spelling changes.
- Use professional security tools for real email protection.
- Report suspicious messages to the appropriate administrator.