Build a Tiny Paper Decision Tree

No adult supervision needed: paper and pens only.

Protective equipment and “do not substitute” warnings have not been recorded for this experiment yet. They are added when its safety review is completed; until then, follow the supervision notes above.

Disposal: Recycle the paper and cards.

Materials

• A large sheet of paper
• About 16 small pieces of card
• A pen

Steps

1. Write one animal on each of 12 cards: for example sparrow, eagle, duck, cat, dog, horse, salmon, goldfish, shark, bee, ant, butterfly.
2. Sort them into four groups: birds, mammals, fish and insects.
3. On the big sheet, write one yes/no question at the top, such as 'Does it have feathers?', and draw a 'yes' branch and a 'no' branch.
4. Keep adding yes/no questions on the branches until every card ends up in a box containing only one group. That is your decision tree.
5. Write 4 new test animals that were not in your 12, such as bat, penguin, whale and spider. Put each through your tree and see where it lands.

What you should see

• Three or four questions are enough to sort all 12 training cards.
• Some test animals land in the wrong box: for example a question like 'Can it fly?' puts a bat with the birds.
• Questions about features that really define a group (feathers, fur, gills) sort new animals better than questions about what they do.

Why it works

A decision tree makes a prediction by asking a series of questions, called conditions, arranged in a hierarchy. Each example starts at the top and follows the answers down a branch until it reaches a leaf, which gives the prediction.

In machine learning the questions are not written by hand: an algorithm chooses them from training examples. Like your paper tree, a tree that fits its training examples perfectly can still be wrong about new examples if its questions do not capture what really matters.

Learn the ideas behind it

• AI basics
• Machine learning

Sources

• Decision trees (Decision Forests course) — Google for Developers