Player dashboard
Progress is stored in your browser. No account is required.
Classification arena
Choose the species
Model training laboratory
Adjust data quality and simulated model resources before training.
Feature selection
Training status
Performance analytics
Charts update after predictions and model training.
Dataset and custom animal studio
Dataset controls
Create a fictional animal
Session records
Review decisions, confidence, timing, and outcomes.
| Round | Actual | Prediction | Confidence | Correct | Time | Model |
|---|---|---|---|---|---|---|
| No classifications yet. | ||||||
Achievements
Species reference guide
Use this guide outside active challenge rounds.
How to play and learn
Game workflow
- Choose a game mode, difficulty, and classifier.
- Inspect visible habitat, diet, movement, size, and body clues.
- Select a species and submit your prediction.
- Review confidence, feature importance, and model feedback.
- Train again with different features and data settings.
Learning goals
Classification uses patterns from labelled examples. Strong features separate similar classes. Noisy or missing clues can reduce confidence.
Accuracy alone may hide weak minority-class performance. Compare precision, recall, F1 score, and the confusion matrix. Use balanced accuracy when classes differ greatly.
Frequently asked questions
Is this a real machine learning model?
The game uses transparent browser-based simulation. It demonstrates classification concepts without requiring a server-side training framework.
What makes a feature informative?
An informative feature helps distinguish one species from others. Habitat, body covering, movement, and diet often work best together.
Why can model confidence be wrong?
Confidence reflects the simulated model score, not certainty. Noise, class imbalance, missing values, and unfamiliar combinations can produce overconfident errors.
What does the confusion matrix show?
Rows represent actual species and columns represent predictions. Diagonal cells are correct classifications. Other cells reveal commonly confused species.
How does feature selection affect performance?
Removing irrelevant clues can reduce noise and training cost. Removing important clues can make related species difficult to separate.
What is a generalisation gap?
It is the difference between training and test performance. A large gap can indicate overfitting or data mismatch.
Why use several evaluation metrics?
Different metrics reveal different mistakes. Precision focuses on false positives, while recall focuses on missed members of a species.
Can I upload my own data?
Yes. Use CSV columns matching the sample file. Uploaded records remain in the current browser session.
Does the game save progress?
Scores, achievements, settings, and discovered species are saved through local storage. Reset progress clears those saved values.