What a predictive model is

Step 1 Warm-up recall

Think of a forecast you rely on, such as a weather forecast. Write what information it uses and what it predicts.

Your answer

Private: nothing you type here is sent or stored.

Step 2 Learning goal

By the end of this lesson you can:

• define a predictive model (Foundation);
• distinguish a rule written by hand from a rule learned from examples (Core);
• name a prediction task and its inputs and output (Core).

Step 3 A familiar example

A letter sorter learns that an address with a certain postcode goes in a certain bag. Nobody wrote a rule for every postcode: the sorter picked it up from many examples.

Step 4 Clear explanation

A predictive model is a rule that takes known information, called features, and returns an estimate of something not yet known, the prediction. Some rules are written by hand: “if the temperature is below zero, expect ice”. In machine learning the rule is learned from examples: past cases where both the features and the answer are known.

Learning from labelled examples is called supervised learning, because each example comes with the right answer, which supervises the learner. The learned rule is then used on new cases whose answer is not known. See the encyclopedia entry on machine learning, and the entry on inference for how a trained model is applied.

Two things are always true. The model can only learn from the examples it is given, so it reflects them, including their gaps. And its predictions are estimates, not facts: the aim is to be right more often than a simple alternative, not to be certain.

Step 5 Worked example (4 marks)

A bank wants to predict whether a loan will be repaid. Describe the task.

1. Examples: past loans whose outcome is known. (1 mark)
2. Features: for instance the amount, the term and the applicant’s income. (1 mark)
3. Prediction: repaid or not repaid for a new application. (1 mark)
4. It is supervised, because each past example has a known answer. (1 mark)

Step 6 Learner practice (instant-marked)

Step 7 Recap

• A model maps known features to an estimate.
• Supervised learning uses examples with known answers.
• Predictions are estimates and reflect the training examples.

Step 8 Exit ticket (ungraded)

Name one prediction task in your own field, its features and what is predicted.

Your answer

Private: nothing you type here is sent or stored.

Step 9 Go further (Extension)

Read the encyclopedia entry on machine learning and find one more kind of learning besides supervised.