Fit / Load
When to use this:
Categorical targets with clear if/then rules.
Fit ID3 Presets are quick starting points; you can still tweak knobs.
Max depth
Sets the maximum number of question layers. Deeper trees fit tighter but can overfit.
Why this matters: Max depth
Deep trees can memorize quirks in the training data.
Shallow trees may miss real patterns.
Use this to show overfitting vs underfitting.
Min samples
Smallest group size allowed to split. Bigger values make the tree steadier.
Why this matters: Min samples
Small groups are noisy and unstable.
Larger minimums create steadier splits.
Too large can hide useful structure.
Numeric bins
Turns numbers into ranges so ID3 can ask questions. More bins means more possible splits.
Why this matters: Bins
More bins create more split options.
Too many bins can chase noise.
Too few bins can blur real differences.
After you click Fit
Start with Findings for the main story.
Use the tree to see the exact questions.
Check accuracy and the confusion matrix.
Findings
What the model saw
Notes
ID3 expects a categorical target. Numeric predictors are binned
into quantiles.
Fit / Load
When to use this:
Numeric targets you want to predict.
Presets are quick starting points; you can still tweak knobs.
Max depth
Sets the maximum number of question layers. Deeper trees fit tighter but can overfit.
Why this matters: Max depth
Deeper trees fit training data more closely.
Shallow trees may miss real patterns.
Use this to show underfit vs overfit.
Min split
Smallest group size allowed to split. Bigger values avoid noisy splits.
Why this matters: Min split
Small groups are noisy and unstable.
Larger minimums create steadier splits.
Too large can hide useful structure.
Complexity (cp)
Prunes weak splits to keep the tree simpler. Higher values prune more.
Why this matters: Complexity (cp)
Higher values prune weak splits.
Lower values allow more detail.
Tuning cp balances fit vs simplicity.
Fit regression tree
After you click Fit
Read Findings for the headline result.
Use the tree to see split points.
Check RMSE and R^2 to judge fit.
Findings
What the model saw
Fit / Load
When to use this:
Categorical targets with numeric predictors.
Presets are quick starting points; you can still tweak knobs.
How we score a split. Gini and information are two common choices.
Max depth
Sets the maximum number of question layers. Deeper trees fit tighter but can overfit.
Why this matters: Max depth
Deep trees can memorize quirks in the training data.
Shallow trees may miss real patterns.
Use this to show overfitting vs underfitting.
Min split
Smallest group size allowed to split. Bigger values avoid noisy splits.
Why this matters: Min split
Small groups are noisy and unstable.
Larger minimums create steadier splits.
Too large can hide useful structure.
Complexity (cp)
Prunes weak splits to keep the tree simpler. Higher values prune more.
Why this matters: Complexity (cp)
Higher values prune weak splits.
Lower values allow more detail.
Tuning cp balances fit vs simplicity.
Fit CART
After you click Fit
Read Findings for the headline result.
Use the tree to see the split rules.
Check accuracy and the confusion matrix.
Findings
What the model saw
Notes
This module fits a CART classification tree (rpart).
Use it for categorical targets.
A decision tree is a flowchart of simple questions.