Glossary

All the key words you need to know for this unit, in alphabetical order.

Accuracy

How often the model gets the right answer. Calculated as: (number correct ÷ total tested) × 100. An accuracy of 90% means the model got 9 out of 10 predictions right.

First used in: Lesson 3

Balanced Dataset

A dataset where each class has a similar number of examples and the data is fair. For example, 10 circles and 10 triangles (not 50 circles and 2 triangles).

First used in: Lesson 2

Bias

When a model learns the wrong pattern from its training data, leading to unfair or incorrect predictions. For example, learning background colour instead of shape.

First used in: Lesson 2

Class / Label

A category or group that the model can predict. In a shapes model, the classes might be "Circle", "Triangle", and "Square". Each class needs its own set of training images.

First used in: Lesson 1

Confidence

A percentage showing how sure the model is about its prediction. 95% = very confident. 55% = not sure. The higher the confidence, the more reliable the prediction.

First used in: Lesson 1

Confusion

When the model mixes up two or more classes — for example, predicting "square" when the image is actually a triangle. This often happens when classes look similar.

First used in: Lesson 4

Ethics

Thinking about what is right and wrong when it comes to using technology and AI. For example: should AI be used to make decisions about people's lives?

First used in: Lesson 6

Evaluation

Judging how well something works by looking at its strengths, weaknesses, and how it could be improved.

First used in: Lesson 5

Fairness

Making sure an AI system treats all groups and inputs equally and doesn't discriminate based on irrelevant factors.

First used in: Lesson 2

Machine Learning (ML)

A type of artificial intelligence where the computer learns from examples (data) instead of being given exact rules. The more examples it sees, the better it gets at making predictions.

First used in: Lesson 1

Model

The "brain" that the computer builds after learning from training data. Once trained, the model can look at new images and make predictions about what they are.

First used in: Lesson 1

Multi-class

A classification task with more than two categories. For example, a model that can recognise circles, triangles, AND squares has three classes.

First used in: Lesson 4

Overfitting

When a model memorises its training data instead of truly learning the patterns. An overfitting model works great on training images but fails on new images it hasn't seen before.

First used in: Lesson 3

Prediction

The model's guess about what it sees in a new image. For example, "I think this is a Circle with 92% confidence."

First used in: Lesson 1

Prototype

An early version of a product, built to test if the idea works before making a full version.

First used in: Lesson 5

Testing Data (Test Data)

Images that the model has never seen before. You use test data to check how well the model works on new inputs. Test data must be separate from training data.

First used in: Lesson 3

Training Data

The example images you give to the model so it can learn. The model studies these images during the training process.

First used in: Lesson 1