Reading 1: What Is a Decision Tree?

You Already Know How to Use One

Before we define a decision tree formally, consider what a school counselor actually does when deciding which students to flag for additional reading support. She does not consider every student equally. She has learned, from years of experience, that certain patterns predict struggle: a reading fluency score below benchmark, combined with teacher concern, is a strong signal. A student who is below benchmark but has high attendance and no teacher concern is in a different category. A student above benchmark rarely needs referral regardless of other factors.

Without knowing it, the counselor is following a decision tree — a series of questions whose answers branch into different paths, leading eventually to a recommendation. The tree exists in her professional judgment, built up from experience. A decision tree in AI makes that structure explicit, formal, and automatic.

The Structure of a Decision Tree

A decision tree is a flowchart-like structure that classifies an input by asking a series of questions and following different paths depending on the answers. It has four kinds of components:

• The root node — the first question asked, at the top of the tree. Every classification starts here.
• Internal nodes — subsequent questions further down the tree. Each internal node tests one attribute of the input.
• Branches — the paths connecting nodes. Each branch represents one possible answer to the question at the node above it.
• Leaf nodes — the endpoints at the bottom of the tree, where no more questions are asked. Each leaf node represents a classification — the tree's conclusion for any input that reaches it.

To classify a new input, you start at the root, answer the question there, follow the appropriate branch, answer the next question, follow the next branch, and continue until you reach a leaf node. Whatever classification is written at that leaf is the tree's answer.

A Worked Example: Student Reading Support

Here is a simple decision tree for deciding whether a student should be referred for reading support, scheduled for monitoring, or left with no action. The tree uses three attributes: reading fluency score (at or above benchmark vs. below), teacher concern (flagged or not), and recent attendance (90% or above vs. below).

The tree has one root node (reading fluency), two internal nodes (teacher concern and attendance), and four leaf nodes (the four possible outcomes). Let us trace it for two different students.

Tracing Student A

Student A has a reading fluency score below benchmark, and her teacher has flagged a concern.

1. Root node: Is reading fluency below benchmark? Yes. Follow the left branch.
2. Internal node: Has the teacher flagged a concern? Yes. Follow the left branch.
3. Leaf node reached: Refer for reading support.

Tracing Student B

Student B has a reading fluency score above benchmark, but his attendance has been below 90% recently.

1. Root node: Is reading fluency below benchmark? No. Follow the right branch.
2. Internal node: Is attendance below 90%? Yes. Follow the left branch.
3. Leaf node reached: Monitor — check in monthly.

Notice that the tree reached a different leaf for Student B than it would have for a student above benchmark with good attendance (who would reach "No action needed"). The attendance question only gets asked for students who are already above the fluency benchmark — it is a secondary signal for students who are otherwise doing well but showing a warning sign.

Why Decision Trees Are Useful in AI

Decision trees have a property that most AI systems lack: they show their work. Every classification comes with a traceable path — a sequence of questions and answers that explains exactly why the tree reached its conclusion. If the tree refers Student A for reading support, you can follow the path and see why: below benchmark fluency, plus teacher concern. You do not have to trust a black box.

This transparency matters enormously in high-stakes domains. A loan approval decision, a medical diagnosis, a student flagging system — these are contexts where "the algorithm said so" is not an acceptable explanation. Decision trees are one of the few AI structures that can provide a human-readable reason for every decision they make.

They also have a property that makes them ideal for teaching: they can be drawn on paper, traced with a pencil, and understood without any mathematics. The concepts involved — asking questions, following branches, reaching a conclusion — are accessible to students from elementary school upward. In Reading 3, we will look at how decision trees can serve as a teaching tool at different grade levels.

One Thing to Notice

Look again at the reading support tree. It makes a recommendation for every student who passes through it, based only on three attributes. A real counselor would consider far more: how long has the student been below benchmark? Is this a new pattern or a persistent one? Are there other factors — a difficult home situation, a recent illness — that the data does not capture?

The tree does not know about any of that. It can only ask about the attributes it was given. Whatever information is not in the tree is invisible to it. This is one of the fundamental limits of decision trees — and of AI classification systems in general. The quality of the classification is bounded by the quality and completeness of the attributes the system can see.

In Reading 2, we turn to how decision trees are built: not by a counselor writing down her intuitions, but by a learning algorithm that examines hundreds or thousands of past examples and figures out which questions to ask. That shift — from hand-crafted rules to learned patterns — is the heart of what makes modern AI different from older rule-based systems.