You see a picture of a fruit and ask, “What fruit is this?” An answer might be Apple, Banana, or Orange. Predicting which named group the picture belongs to is a classification task.
A category is a named group, also called a class. Classification means predicting the class for an input. A model that does this is called a classifier.
Why do we need classification?
Recognizing one apple is easy for a person. Sorting thousands of new pictures or messages takes time. A trained classifier can suggest a category for each new item, helping a system sort it or send it to the right place.
Here are three practical uses, with example predictions.
- Email filtering. Read an incoming email and predict Spam or Not spam. The email service can use that category to sort unwanted messages into the spam folder.
- Support routing. Read a customer’s message and predict Billing, Delivery, or Returns. The category helps send the request to the matching support team.
- Language detection. Read a sentence and predict its language, such as English, Hindi, or Spanish. This can help a translation tool choose the input language.
This is useful when people can provide examples with known categories, but writing a dependable rule for every new item would be difficult. The prediction supplies a category that a system can act on.
Learn from pictures with known names
Training examples pair fruit pictures with their known names: an apple picture with Apple, a banana picture with Banana, and an orange picture with Orange. Each known name is the label.
The model learns patterns from those labeled examples. Later, we give it a new picture without the name. It predicts a category, such as Apple. The drawing below illustrates this task; an image model is not running on the page.
The new picture matters
An apple picture and a banana picture should lead to different category predictions. The purpose is to recognize the new input using what was learned from examples.
A classifier can make mistakes. If it predicts Orange for a banana, compare that prediction with the known label, Banana, to see the mistake. The methods used to learn these patterns receive their own later chapters.
Choose the fruit category
Try the recognition task yourself. Pick a picture, then choose its category.
What fruit is in this picture?
Choose one category.
This activity checks your answer against the known label. It does not run a trained image model.
A category name can be a digit
Suppose the input is a picture of a handwritten 7. The task is to identify which digit was written. The possible category names are 0 through 9.
A prediction of 7 names the digit in the picture. It is still classification. Category names can be words or numeric codes.
Classification in one infographic
Labeled pictures, a new input, and a predicted category.
Practice code and sources
Run the recognition practice
Optional: Node.js 20 or newer, no extra packages. Copy the full source into classification-example.mjs. This checks a practice choice against its known answer, without training an image model.
node classification-example.mjsCorrect category.
Known answer: BananaChange pictureId and myAnswer near the bottom, then run again.
Full source
// A recognition practice activity, not a trained image model.
export const FRUIT_CATEGORIES = Object.freeze(["Apple", "Banana", "Orange"]);
export const FRUIT_PICTURES = Object.freeze([
{ id: "apple", label: "Apple picture", answer: "Apple" },
{ id: "banana", label: "Banana picture", answer: "Banana" },
{ id: "orange", label: "Orange picture", answer: "Orange" },
]);
export function checkFruitChoice(pictureId, answer) {
const picture = FRUIT_PICTURES.find(item => item.id === pictureId);
if (!picture || !FRUIT_CATEGORIES.includes(answer)) throw new RangeError("Choose an available picture and category.");
return { correct: answer === picture.answer, answer: picture.answer };
}
// This checks a practice answer. It does not train or run an image model.
// Try apple, banana, or orange, then choose Apple, Banana, or Orange.
const pictureId = "banana";
const myAnswer = "Banana";
const result = checkFruitChoice(pictureId, myAnswer);
console.log(result.correct ? "Correct category." : "Try another category.");
console.log("Known answer: " + result.answer);
Sources and teaching notes
The definition follows Google’s classification definition. Google’s supervised-learning introduction explains input-label pairs and prediction on new inputs. The fruit drawings and predictions are illustrative. The practice activity uses a fixed answer key.
Anthropic’s support-ticket routing guide describes using classification to route customer requests. The practical examples above use illustrative category predictions.
← Chapter 2: Regression
Next: inputs, features, and known answers. Chapter 4 is planned.
