How Machines Learn

🧠 What a model is

When AI learns, it builds a model: a set of rules or numbers that turns inputs (features) into a prediction. A model for crop prices might take the season and the rainfall, and predict the price of maize next month. The model is the part that is "trained".

🌳 Decision trees

A decision tree is one simple kind of model. It asks yes/no questions, one after another, until it reaches an answer: "Does it have feathers? Yes → bird. No → Does it breathe with gills? Yes → fish. No → mammal." Training means finding which questions to ask, and in what order.

🔁 Training is trial and improvement

Most models start bad and improve. They make predictions on the training data, measure how wrong they are, adjust a little, and repeat, often thousands or millions of times. Neural networks, the models behind modern speech and image AI, learn this way by adjusting millions of numbers.

📈 Accuracy and testing

Accuracy is the share of predictions that are correct. A model that is right 90 times out of 100 has 90% accuracy. It must be measured on test data the model has never seen. A model can score 100% on its training data and still fail on new examples.

🎯 Overfitting

Overfitting is when a model memorises its training examples instead of learning the general pattern, like a learner who memorises past-paper answers but can't solve a new question. In the explainer above, a tree can sort every training animal perfectly and still get a dolphin wrong.

Key points

• A trained model turns inputs into a prediction.
• A decision tree asks yes or no questions until it reaches an answer.
• Training is trial and improvement, repeated many times.
• Neural networks learn by adjusting millions of numbers.
• Accuracy is the share of predictions that are correct.
• Always measure accuracy on test data the model hasn't seen.
• Overfitting is memorising the examples instead of the pattern.

Try it for real: Train and test your tree

• In the explainer, build a tree that sorts all the training animals, then test it on the new ones. Note how many it gets right.
• Start again and build a different tree. Which questions worked best on the new animals, and why?