| Class | Colour | Weight | Length | Width | Shape | Texture |
|---|
| Round | Actual | Player | Model | Confidence | Points |
|---|
How classification works
A classifier learns patterns linking fruit features to labels. It estimates which known class best matches a new sample.
Training examples build the model. Testing examples measure whether those patterns generalise beyond memorised records.
Features and targets
Colour, weight, shape, size, and texture are input features. The fruit name is the target label.
Useful features separate classes consistently. Weak or noisy features can reduce confidence and increase errors.
Confidence and unknowns
Confidence represents model certainty, not guaranteed truth. Similar fruits may receive close probabilities.
The confidence threshold enables an unknown prediction. This prevents forced guesses when evidence is weak.
Evaluation guide
Accuracy measures total correctness. Precision and recall examine class-specific errors from different perspectives.
The confusion matrix reveals repeated mix-ups. Feature importance estimates which measurements influenced decisions most strongly.
Model descriptions
Finds similar training samples and lets nearby labels vote.
Uses readable feature rules and branching thresholds.
Combines feature likelihoods with class frequencies.
Builds weighted class scores from numerical features.
Averages many noisy rule-based trees.
Uses layered weighted transformations and nonlinear activations.