Flow, process and networks
Decision tree
Shows decision rules or fitted model splits leading to outcomes or predictions.
- Diagram
- Decision logic
- Prediction
What it answers
Which rules lead to each outcome?
- Shape
- Rule branches ending in outcomes
- What it shows
- Decision paths and segmentation.
Example
The question
Which customers are likely to cancel, by simple rules?
Rules for spotting customers likely to cancel (decision tree)
Illustrative data. It belongs to no company.
When to use it
Use it when
Explaining authored decisions or an interpretable fitted tree.
Avoid it when
When a tree implies causal rules or omits validation.
How to read it
Follow branch conditions from root to a terminal node.
Note
Distinguish expert-authored rules from fitted models.
The data you need
- Layout
- Rule-node table
- Fields
- NodeID: text; ParentID: text/blank; RuleOrOutcome: text; NodeType: category
- One row is
- One decision or outcome node. What is the grain?
- Rules
- Label branches with mutually understood conditions. Ensure parent-child connections and terminal outcomes are valid.
- Optional
- BranchLabel,Probability,Prediction. Fitted models need suitable training data and a training step.
- Preparation
- Validate unique node IDs, parent links, roots and absence of cycles. For size-encoded hierarchies, state whether totals are leaf-only or inclusive.
- Units
- Tree levels and branch position are structural. Size-encoded components use a named value unit; parent shares divide by the chosen parent total.
Worked example
Rule income>50000 leadsApprove; otherwiseReview. Income 60000 followsApprove. This is an authored rule, not evidence of model accuracy.
| NodeID | ParentID | RuleOrOutcome | NodeType |
|---|---|---|---|
| n1 | Income > 50000 | Decision | |
| n2 | n1 | Approve | Outcome |
| n3 | n1 | Review | Outcome |