The Akinator

Remember that game "20 Questions"? "Is it a person?" -> Yes. "Is he alive?" -> No. "Is he a scientist?" -> Yes. "Is it Einstein?" -> Yes.

That is literally a Decision Tree.

The Flowchart of Life

A Decision Tree is just a flowchart that splits your data into smaller and smaller groups until it's confident enough to guess.

Imagine you are hiring a developer:

1. Q1: Do they know Python?
   • No: Don't hire.
   • Yes: Go to Q2.
2. Q2: Do they copy-paste from StackOverflow?
   • No: Suspicious. Don't hire.
   • Yes: HIRE IMMEDIATELY.

Interactive Demo: The Splitter

The tree makes "Cuts" in the data.

• Cut 1: Is Color == Red?
• Cut 2: Is Size > 5cm?

Unlike the Linear Regression line which is diagonal, Decision Tree cuts are always horizontal or vertical. They are like steps on a staircase.

The Code (Python)

We use DecisionTreeClassifier. It's great because you can actually see why it made a decision (unlike Neural Networks which are black boxes of magic).

from sklearn.tree import DecisionTreeClassifier
 
# 1. The Question Asker
model = DecisionTreeClassifier(max_depth=3)
 
# 2. Train on fruit data
# Features: [Redness (0-10), Size (0-10)]
X = [[10, 5], [1, 5], [10, 1]]
y = ["Apple", "Orange", "Cherry"]
model.fit(X, y)
 
# 3. Predict
print(model.predict([[9, 6]])) 
# Output: "Apple"

The Catch (Overfitting)

If you let the tree ask infinite questions, it becomes too specific. "Is it a round fruit, red, with a scar on the left side, bought on a Tuesday?" -> "That specific Apple I ate yesterday."

This is called Overfitting. The tree memorized the data instead of learning the rules. We stop this by setting a max_depth (limiting the number of questions).

Summary

Decision Trees are simple, interpretable, and work exactly like your brain when choosing what to eat for lunch. "Is it healthy? -> No. -> Is it tasty? -> Yes. -> Pizza."

Next up: What happens when we plant a whole forest of these trees?