Learn AI /How Machines Learn /Prediction
Prediction
Also called: output, model output.
Back to the milestone you came from
In five seconds
What it is
The result a model produces from input, such as a class, number, ranking or next token.
In plain words
A prediction is the result a model gives after it receives input.
An everyday example
At an admissions desk, staff receive many applications every hour. A clerk quickly looks at each form and places it into one of three trays: likely complete, likely missing documents, or needs human review. The clerk uses visible clues such as blank fields, missing signatures and whether required attachments are listed. The tray chosen for each new form is like the model’s prediction.
A human clerk can explain their reasons in ordinary language and may apply written rules directly. Many machine-learning models do not work by following a short checklist, and their internal processing can be harder to interpret. Also, some model predictions are numbers or probabilities rather than neat tray labels.
Remember thisA prediction is the output a trained model produces for an input. It may be a category, number, ranking or next token, and it may be wrong.
Understand it
Understand it
Why it matters
If you use or manage AI tools, prediction is the point where the system affects action. A predicted label may move a case into a queue. A predicted score may influence priority. A predicted next word may shape a draft. Knowing that a prediction is an output, not a guarantee, helps you ask better questions: What kind of output is this? How often is it wrong? What happens if it is wrong? Who should review it? This matters for fairness, safety, quality and accountability.
From input to prediction
After viewing the diagram, a reader should be able to say that a prediction is the output produced when a model processes new input.
01
1. Input arrives
New data is provided to the model, such as a form, image, sentence or table row.
Draw a box labelled 'Input' on the left with four small icons underneath: document, photo, text lines and mini table.
02
2. Features are read
The model uses parts of the input that help it detect patterns. These parts are often called features.
Draw a middle box labelled 'Model' with small highlighted snippets entering it from the input box to suggest selected information.
03
3. Output is produced
The model returns a result. Depending on the task, this might be a category, a number, a ranking or the next token.
Draw a right box labelled 'Prediction' with four example chips: 'complete', '7.2', 'A > B > C', and 'next word'.
04
4. Check if needed
The prediction can later be compared with the correct outcome or used by a person to support action.
Add a smaller dashed box below the prediction box labelled 'Human check / later outcome' connected by a dashed arrow.
A left-to-right flow diagram with three boxes connected by arrows: input, model processing, and prediction. Small callouts show that the same kind of system can output a class label, a number, a ranking or a next token. Read from left to right.
Hovering or tapping each example chip reveals a short note: category for classification, number for regression, ranking for retrieval or recommendation, and next token for language generation.
Describe this without the diagram
The diagram shows a simple left-to-right process. First, new input enters the system, such as a document, image, text or table row. Second, the model processes useful parts of that input, often called features. Third, it produces a prediction. The prediction may be a category like 'complete', a number like '7.2', a ranked list, or the next token in a sequence. A dashed note underneath says that a person can review the result or compare it later with the correct outcome. The main point is that prediction means the model’s output on new input.
With reduced motion: A static three-box flow is shown: Input, Model, Prediction, with example outputs listed in the final box and a dashed note below for human check or later outcome.
A prediction is what a model produces after you give it input. The input might be a form, an image, a sentence or a table of numbers. The output could be a category, a score, a number, a ranked list or the next word in a sequence.
You can think of it as the model’s best estimate based on patterns it learned from examples. When run, the model processes input features and generates an output. A feature is a piece of information used by the model, such as the number of missing fields in an application.
A prediction is not automatically the same as a label or ground truth. A label is the expected answer attached to an example so a model can learn. Ground truth is the checked reference answer used for training or evaluation. In a classification task, the prediction may instead be a predicted category such as "complete" or "incomplete". In other tasks it may be a number, such as a predicted delivery time.
People often confuse prediction with certainty. Many models can also produce a probability or confidence-like score, but that still does not guarantee the result is right. Predictions can help people sort, prioritise or estimate, but they still need checking in the right context.
How “Sorting applications into trays” maps onto the real thing
| In the story | In the AI system |
|---|---|
| A new application arriving at the desk | New input given to the model |
| Visible clues on the form, such as blank fields or missing signatures | Features used to produce the output |
| The tray chosen by the clerk | The prediction or model output |
| Later checking the file against the requirements list | Comparing the prediction with ground truth or evaluation results |
A worked example
A political affairs officer must identify sections of an implementing partner’s progress report that need follow-up before the file enters clearance.
Starting from: The activities section says delivery is complete, consultation with national counterparts is pending, and the supporting annex is not attached.
- The approved model receives the report text and available document indicators as input.
- It maps those features to scores for the categories ready for clearance, follow-up needed and manual review.
- The category with the highest score is returned as the prediction.
- The officer checks the underlying text and annexes before deciding what action to take.
Result: Prediction: follow-up needed. The officer confirms that consultation remains pending and the annex is absent, then requests clarification from the implementing partner.
In a United Nations workplace
Before close of business, a political affairs officer in an inter-agency secretariat reviews an implementing partner’s progress report using an approved internal tool. The tool predicts whether each section is ready for clearance, needs follow-up or requires manual review.
Each category returned is a prediction based on patterns the model learned from earlier inputs. It is not the report’s label, a finding about the partner’s performance or a clearance decision.
Before you try thisThe officer must verify the report and supporting material rather than act on the category alone. Personal, confidential or security-sensitive material must not be entered into any tool that has not been approved to process it.
Where you may hear it
- A vendor demonstration during a procurement meeting
- A progress report review before front-office clearance
- A model-validation session in an evaluation office
- An inter-agency working group discussing a forecasting dashboard
Avoid confusion
Avoid confusion
Often confused with
Label
The expected answer attached to an example so a model can learn what output should match that input.
Ground truth
The checked reference answer used to train or evaluate a model, so its output can be compared with the best available real-world answer.
Common misconceptions
Common misconception: A prediction means the model is telling the future.
Correction: Sometimes the output is about the future, such as tomorrow’s demand, but often it is not. A model can predict a present category like 'spam', a missing value, a translation token, or which document is most relevant. In machine learning, prediction means the produced output more broadly.
Common misconception: A prediction is the same as the correct answer.
Correction: A prediction is only the model’s output. The correct answer may be unknown at the time, or may be checked later. That is why evaluation compares predictions with verified outcomes rather than assuming they match.
Common misconception: If a model gives a very high score, the prediction must be reliable.
Correction: A high score can mean the model strongly favours one output, but strength of preference is not the same as truth. Some models are poorly calibrated, meaning their scores do not match real-world accuracy very well.
Quick check
Which statement best describes a prediction in machine learning?
A prediction is the model’s produced output, such as a class, score or next token. It is not automatically the verified correct answer, and it is not the training procedure or a hand-written rule list.
Go deeper
Go deeper
The simple version
A prediction is the result a model gives after it receives input.
The technical version
In machine learning, a prediction is an output generated during inference when a trained model maps input features to an output for its task. It may be discrete, as in classification; continuous, as in regression; a ranked set of items; a cluster assignment; or the next token in sequence modelling. The prediction may be returned directly as a category or value, or selected from scores, logits or probabilities. In supervised settings, it is distinct from the label attached to an example and from ground truth, the checked reference used for training or evaluation.
Reviewed external resources
EncyclopaediaPrediction (opens in a new tab)Wikipedia
This Wikipedia link gives a broad overview of prediction as a general concept and suits readers who want a starting point with related terms.
Intermediatelink checked 2026-07-31
CourseMachine Learning Crash Course (opens in a new tab)Google
This Google course link is useful for beginners who want practical machine-learning context, including how models produce outputs from inputs.
Intermediatelink checked 2026-07-31
VideoStatQuest (opens in a new tab)StatQuest with Josh Starmer
This StatQuest video resource suits learners who prefer clear spoken explanations with simple visuals and step-by-step intuition.
Starterlink checked 2026-07-31
VideoVideo explanations: Prediction (opens in a new tab)YouTube search
This YouTube search results page helps readers find different video explanations of prediction and suits people who want to compare teaching styles.
search results, not one pagelink checked 2026-07-31
DiscussionCommunity discussion: Prediction (opens in a new tab)Reddit search
This Reddit search results page helps readers see community discussions and common points of confusion, and suits readers who want informal explanations with caution about reliability.
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
A prediction is the output a trained model produces for an input. It may be a category, number, ranking or next token, and it may be wrong.