Interactive machine learning game
Investigate what drives a tree model
Build a simulated model, rank predictors, detect leakage, compare methods, and explain individual decisions.
Player status
Ready
Score
0
Level
1
Streak
0
Hints
3
Level progress0 / 500 XP
1. Scenario
Choose a dataset
Time remaining60s
2. Model laboratory
Configure the tree model
3. Feature workshop
Select and inspect features
Select a feature
Feature details, distributions, missing values, and warning signs appear here.
4. Canvas decision path
Explore how the tree makes a decision
Prediction
—
Confidence
—
Active path
0 nodes
5. Importance dashboard
Compare global and local explanations
Local explanation
Prediction contribution waterfall
Feature relationships
Correlation and interaction view
Evidence table
Detailed feature importance
| Rank | Feature | Type | Impurity | Permutation | Gain | Split count | Local effect | Stability | Flags |
|---|
Model report
Performance and diagnostics
Learning note
Train a model to reveal an explanation about importance bias, correlation, leakage, or stability.
Achievements
Mastery badges
Round history
Recent attempts
| Round | Scenario | Mode | Score | Result |
|---|
Learning guide
Feature importance concepts
Impurity importance rewards features that create large training splits. Permutation importance measures validation damage after shuffling. Correlated features can substitute for each other, so each method may distribute credit differently.
High-cardinality variables can receive inflated split-based scores. Correlated predictors may share or hide importance. Leakage features reveal information unavailable when a real prediction is made.
Global importance summarizes behavior across many records. Local importance explains one prediction. A globally weak feature can still dominate a particular decision when its value is unusual.