Learn AI /How Machines Learn /Classification
Classification
- Starter
- How Machines Learn
- Programme & data teams
Also called: classifier, classify.
Back to the milestone you came from
In five seconds
What it is
A machine learning method that assigns one or more predefined category labels to an input based on learned patterns.
In plain words
Classification is teaching a system to put each new item into one or more named buckets, like sorting messages into “spam” or “not spam”.
An everyday example
Sam is shown many buttons already placed in trays labelled “red”, “blue”, “two-hole” and “four-hole”. From those examples, Sam learns which visible features correspond to each label. For a new button, Sam checks its colour and number of holes, then places it in the matching trays. If only colour labels are allowed, it receives one label. If colour and hole-count labels may both apply, it receives more than one.
A person tuning a radio can use broad real-world knowledge and can change the categories if needed. A classifier usually works only with the classes it was trained for, and its internal matching process is mathematical rather than conscious listening. Also, many classifiers output scores for all classes at once, not a person-like step-by-step judgement.
Remember thisClassification uses patterns learned from labelled examples to assign one or more predefined categories to a new input. A score supports the prediction, but is not proof that the label is correct.
Understand it
Understand it
Why it matters
Classification affects how information is routed, filtered and prioritised. A system may flag content as risky, sort messages into queues, or label documents for faster handling. If you understand classification, you can ask practical questions: What are the classes? Who chose the labels? How often does it confuse one class with another? What happens when it is unsure? These questions matter because errors have consequences. A false alarm can waste time or unfairly block something. A missed case can hide an urgent issue. Non-specialists do not need to build classifiers, but they often need to judge whether one is suitable, what data it needs and where human review is still necessary.
How classification turns an input into a category
After viewing this, the reader should be able to say that classification takes an input, measures useful features, compares learned patterns and outputs one named class.
01
1. New input arrives
A text, image, sound clip or data record is presented to the model for sorting.
Draw a box on the left labelled “Input” with small icons for a document, photo and waveform. Add a caption: “Example: message, image or form entry”.
02
2. Signals are measured
The system represents parts of the input as features, such as words, shapes, colours or numeric values.
Draw a second box labelled “Features” containing tokens like “urgent”, “invoice”, a small shape outline and bars for numeric values.
03
3. Learned patterns are compared
The model uses patterns learned from labelled examples to estimate how strongly the input matches each class.
Draw a third box labelled “Model” with three horizontal score bars named Class A, Class B and Class C, each at different lengths.
04
4. A class is returned
The output is a prediction, often the highest-scoring class, sometimes alongside confidence-like scores.
Draw a final box labelled “Prediction” with one highlighted result such as “Invoice” and smaller secondary scores beneath it.
A left-to-right flow diagram with four boxes connected by arrows. It shows a new input entering the system, being converted into features, compared with patterns learned during training, and ending with a class prediction and confidence scores. Read from left to right.
Hovering or tapping each box reveals one concrete example for text, image and tabular data; arrows highlight in sequence from input to prediction.
Describe this without the diagram
The diagram is a four-step flow from left to right. First, a new input enters, such as a message, image or form entry. Second, the system measures features from that input, like words, shapes or numbers. Third, the model compares those features with patterns learned earlier from labelled examples and produces scores for each possible class. Fourth, it returns the class with the strongest score, for example labelling a document as an invoice. The main lesson is that classification outputs a named category, not an open-ended answer or a numeric estimate.
With reduced motion: A static left-to-right diagram with four labelled boxes: Input, Features, Model and Prediction. Short example text under each box replaces any animated highlighting.
Classification is used when you want a system to choose one or more categories from a fixed set. For example, an email might be labelled “spam” or “not spam”, or a document might be labelled “invoice”, “contract” or “report”. The input can be text, an image, a sound clip or a row of numbers.
During training, the system is shown many examples along with the correct category or categories. Those correct categories are called labels. From these examples, it adjusts internal settings so that certain patterns in the input become linked with certain labels.
Later, during inference, a new input goes in. The system turns that input into measurable signals, often called features, such as words present in a message or shapes in an image. It then produces a prediction: which class or classes apply, sometimes with a score for each class.
Classification is different from regression, which predicts a number such as temperature or price. It is also different from clustering, which groups similar items without using pre-named categories. In classification, the categories are defined in advance, even if the model sometimes gets them wrong.
How “Sorting buttons into labelled trays” maps onto the real thing
| In the story | In the AI system |
|---|---|
| list of station types such as news, music or talk | predefined classes or labels |
| clues like language, voice and music style | features extracted from the input |
| Aisha choosing the best matching station type | the classifier producing a prediction |
| weak or noisy signal | ambiguous input that increases the chance of error |
A worked example
Classifying a facilities ticket before the duty officer hands over the service desk at a Bonn duty station.
Starting from: Ticket text: “The meeting room on level 3 is too warm, and the wall control does not respond.”
- The classifier receives the ticket text as input.
- It identifies features associated with the available labels, including “too warm”, “wall control” and “does not respond”.
- It compares those features with patterns learned from tickets previously labelled “temperature issue”, “equipment fault” or “access issue”.
- It calculates a score for each label: temperature issue 0.94, equipment fault 0.72 and access issue 0.04.
- Because this is a multilabel classifier, it returns every label above the set threshold of 0.50.
Result: Predicted labels: “temperature issue” and “equipment fault”. The ticket can be routed to facilities maintenance, with the duty officer able to correct the labels if the request has been misread.
In a United Nations workplace
Late in the afternoon, a budget officer in a country office is reviewing applications for a finance roster before the proposed shortlist goes to the front office at close of business. An approved screening tool classifies each application as “profile matched”, “manual review” or “required criterion not evidenced”, using patterns learned from previously labelled applications.
Classification determines which predefined label each application receives. The budget officer needs to examine how the training labels were assigned, what evidence drives each category, how borderline cases are handled and whether errors fall unevenly across groups. The output is a triage label, not a recruitment decision.
Before you try thisApplications contain personal and potentially sensitive HR information. They must not be uploaded to an external or unapproved tool. A classifier must not make or effectively determine the shortlist: every exclusion requires meaningful human review, since biased training labels or missed evidence could unfairly affect a candidate.
Where you may hear it
- A vendor demonstration at an ICT governance board meeting
- First reading of requirements for a document-management system
- An evaluation office discussion on coding survey responses
- Clearance of a concept note for an automated triage pilot
Avoid confusion
Avoid confusion
Often confused with
Regression
A supervised learning method that learns from example inputs and known target values to predict a numeric quantity for a new input.
Clustering
A method that groups similar examples together so patterns in data can be found without pre-written categories.
Unsupervised learning
A way of training a model on unlabelled data to identify possible structure, similarities, groups or unusual cases without target answers for each example.
Common misconceptions
Common misconception: Classification always gives a yes-or-no answer.
Correction: Some classifiers are binary, with two classes such as approved or rejected, but many choose among three or more classes. A system can also return a score for each class, not just a single hard label.
Common misconception: If the model is 90% confident, it is correct 90% of the time for every case.
Correction: A score is not automatically a trustworthy probability. Some models are poorly calibrated, meaning their confidence-like numbers do not match real-world accuracy. Performance can also differ across classes and contexts.
Common misconception: Classification finds natural groups in data by itself.
Correction: That describes clustering more than classification. In classification, the categories are named in advance and the system learns from examples linked to those labels.
Quick check
Which situation is a classification task?
Classification chooses from predefined categories. “Urgent”, “routine” and “irrelevant” are named classes decided in advance. Predicting rainfall is regression, grouping unlabeled photos is clustering, and writing a summary is a text generation task.
Go deeper
Go deeper
The simple version
Classification is teaching a system to put each new item into one or more named buckets, like sorting messages into “spam” or “not spam”.
The technical version
In machine learning, classification is a supervised learning task that maps an input x to one or more labels from a finite set of classes. A model is trained on labelled examples to learn a decision function or class score based on features in the input. At inference time, it computes scores for the possible classes and returns one or more labels, depending on the design. Common variants include binary classification, multiclass classification and multilabel classification, where one item can receive more than one label. Evaluation often examines per-class performance, confusion between classes and trade-offs such as false positives versus false negatives.
Reviewed external resources
EncyclopaediaStatistical classification (opens in a new tab)Wikipedia
This Wikipedia link gives a broad overview of statistical classification and suits readers who want a general reference with related concepts and terminology.
Intermediatelink checked 2026-07-31
CourseMachine Learning Crash Course (opens in a new tab)Google
This Google course link suits beginners who want a structured introduction with practical machine learning teaching materials and examples.
Intermediatelink checked 2026-07-31
VideoStatQuest (opens in a new tab)StatQuest with Josh Starmer
This StatQuest video link suits visual learners who want a friendly, step-by-step explanation from an educator rather than a formal text.
Starterlink checked 2026-07-31
VideoVideo explanations: Classification (opens in a new tab)YouTube search
This YouTube search results link helps readers browse multiple video explanations of classification and suits people choosing a teaching style they like; it is a search page, not one authoritative source.
search results, not one pagelink checked 2026-07-31
DiscussionCommunity discussion: Classification (opens in a new tab)Reddit search
This Reddit search results link lets readers see community questions and discussions about classification and suits readers who learn from examples and debate; it is a search page, not one authoritative source.
search results, not one pagelink checked 2026-07-31
Where this came from
Drafted with a language model through an offline pipeline, checked against this site's schema, and every external link above fetched to see whether it answers. Treat it as a starting point and check anything that matters against the primary sources. How this site is built and what that means.
The one thing to remember
Classification uses patterns learned from labelled examples to assign one or more predefined categories to a new input. A score supports the prediction, but is not proof that the label is correct.