Specialty Digest

Global English editionLearners’ Study Dashboard — in development
The Knowledge Library for All

Supervised Learning

nounMachine Learningalso supervised machine learning, learning with a teacher

In one line

Supervised learning is a type of machine learning in which a model learns from labeled examples to predict the correct output for new inputs.

Supervised learning is a type of machine learning in which a model learns from examples that come with the correct answers, called labels, so that it can predict the right output for new, unseen inputs. The word supervised refers to the labels, which act like a teacher marking each example. A model might be shown thousands of emails marked spam or not spam and learn to classify future messages, or be shown houses with their sale prices and learn to estimate the price of a new listing. It is the most widely used form of machine learning in business and science.

How supervised learning works

The process begins with a labeled dataset, in which each example has features, the measurable inputs, and a label, the desired output. A learning algorithm starts with a model whose internal parameters are set to initial values. It makes predictions for the training examples, compares them with the labels using a loss function that measures the error and then adjusts the parameters to reduce the loss. This is repeated over many passes through the data. When training is complete, the model is tested on separate data it has not seen to estimate how well it will perform in practice. If results are good enough, the model is deployed to make predictions on new inputs.

Types of supervised learning problems

Two broad kinds of tasks are most common.

  • Classification: the output is a category. Examples include deciding whether a tumor image is benign or malignant, identifying a spoken word or detecting fraudulent transactions. Problems with two categories are called binary, and those with more are multiclass.
  • Regression: the output is a number. Examples include predicting house prices, electricity demand or the time a delivery will take.

Some tasks combine these ideas, such as predicting several labels at once for the same input, called multilabel classification, or predicting structured outputs such as a sentence in translation.

Common supervised learning algorithms

Many algorithms are used, and the best choice depends on the data and goal.

  • Linear and logistic regression: simple, fast methods that fit a line or a probability curve to the data.
  • Decision trees and random forests: methods that split the data with a series of questions, often combined in ensembles for accuracy.
  • Support vector machines: methods that find the boundary that best separates classes.
  • k-nearest neighbors: a method that predicts from the most similar training examples.
  • Neural networks: flexible models that learn complex patterns, especially in images, speech and text.

In practice, teams often try several algorithms and compare them on validation data before choosing one.

Training, validation and testing

To measure how well a model will work on new data, practitioners divide the labeled data into subsets. The training set is used to fit the model, the validation set is used to compare options and tune settings, and the test set is held back for a final check. A model that does very well on training data but poorly on new data is overfitting, meaning it has memorized details and noise instead of general patterns. A model that does poorly on both is underfitting. Techniques such as cross-validation, regularization and gathering more data help to balance these problems.

Applications of supervised learning

Supervised learning is used in email spam filters, credit scoring, medical image analysis, speech recognition, handwriting recognition, product recommendations, language translation, and weather and demand forecasting. In healthcare, models trained on labeled scans help radiologists detect disease. In finance, models trained on historical transactions flag unusual activity. In manufacturing, models trained on sensor data predict when machines are likely to fail. Because a labeled example links an input directly to a target, this approach tends to give accurate, measurable results when good data are available.

Advantages and disadvantages

The main advantage is clarity. Because the correct answers are known for the training data, progress can be measured with metrics such as accuracy, precision and recall, and results are often easier to evaluate than in other types of learning. The method works for a very wide range of tasks. The main drawback is the need for labeled data, which can be expensive and slow to produce because people often have to annotate examples by hand. Labels can contain mistakes or bias, and a model will learn those faithfully. Models can also fail when real-world data differ from the training data, a problem known as distribution shift.

Supervised vs. unsupervised learning

In unsupervised learning, the data have no labels, and the model looks for structure on its own, such as groups of similar customers. Semi-supervised learning uses a small amount of labeled data together with a large amount of unlabeled data, and self-supervised learning creates labels automatically from the data itself, as when a language model learns to predict the next word. Reinforcement learning differs again, because an agent learns from rewards for its actions and not from labeled examples. Supervised learning is the choice when labeled examples are available and the goal is a specific prediction.

Challenges in supervised learning

Practitioners face several recurring problems. Data quality is the most common. Missing values, inconsistent labels and unrepresentative samples all reduce accuracy, and cleaning and preparing data often takes more time than modeling. Imbalanced classes, such as fraud cases that make up a tiny share of all transactions, can make a model look accurate while missing the cases that matter, so metrics such as precision and recall are used alongside accuracy. Data leakage, where information from the test set accidentally influences training, can produce falsely good results. Finally, models need monitoring after deployment, because patterns in the real world shift and a model that worked last year may drift.

A simple example

Imagine a bank that wants to predict whether a loan applicant will repay. It collects past loans, with features such as income, existing debt and employment length, and a label showing whether each loan was repaid. A supervised algorithm learns the relationship between the features and the outcome. The bank tests the model on loans it did not use for training and checks that errors are acceptable across different groups of applicants. If the results meet its standards, the model becomes one input to lending decisions, with human review for borderline or high-impact cases. The example shows both the power of the approach and the need for care about fairness and oversight.

FAQs

What does labeled data mean? It means each example comes with the correct answer, such as a photo tagged cat, that the model is trained to predict.

Is supervised learning the same as deep learning? No. Deep learning is a family of neural network methods that can be used for supervised, unsupervised or other kinds of learning.

How much labeled data is needed? It depends on the task and model. Simple problems may need hundreds of examples, while complex ones need millions.

Key facts

  • In supervised learning, each training example includes an input and a correct label that the model learns to predict.1
  • The two main supervised tasks are classification, which predicts a category, and regression, which predicts a number.2
  • Performance is estimated on data held out from training to check that the model generalizes.2

Related concepts

Quick checkWhat are the two main types of supervised learning problems?Show answer

Classification, which predicts a category such as spam or not spam, and regression, which predicts a continuous number such as a price.

Sources

  1. IBM, What is supervised learning?, https://www.ibm.com/think/topics/supervised-learning
  2. scikit-learn, Supervised learning, https://scikit-learn.org/stable/supervised_learning.html
  3. Google, Machine Learning Glossary, https://developers.google.com/machine-learning/glossary

Editorially reviewed by Specialty Digest Editorial TeamLast reviewed November 17, 2026Researched and drafted with AI assistanceReport an issue