How models work

What machine learning really is

Machine learning is teaching a computer to find patterns in data, then use those patterns to make predictions about new data it has never seen.

Imagine you want to predict a student's exam score. You have records of many students: how many hours they studied, their attendance, their score in a previous test, and the exam score they finally earned. A machine learning model studies those records, learns how these things relate to the final score, and can then estimate the score of a student it has never seen.

Some words you will use the whole way through:

• The records you learn from are the training data.
• Each input (study hours, attendance, previous score) is a feature.
• The thing you want to predict (the exam score) is the target.

A simple but powerful model is the decision tree. It splits the data with yes or no questions, like "did the student study more than 5 hours?", until it reaches a sensible prediction at the end of each branch. Ask enough good questions and you get a good estimate.

You do not need heavy maths to begin. You need data, a model, and four steps you will practice in this course: define, fit, predict, evaluate.

Practice

Think of something you would like to predict: exam scores, crop yield, shop sales. Write down the target, and three features that might help predict it. You will reuse this way of thinking throughout the course.