Machine learning is a way of building computer systems that learn from data. Instead of a programmer writing every rule, the computer is shown examples and works out patterns by itself, then uses them to make predictions or decisions about new cases. It powers spam filters, product recommendations, voice recognition, fraud alerts and many medical tools. This guide answers common questions about how machine learning works and what it is used for, and links to detailed articles. The main entry is machine learning.
What is machine learning?
Tom Mitchell’s 1997 textbook gave a classic definition: a program learns if its performance on a task improves with experience [2]. The term was popularized earlier by Arthur Samuel, whose checkers program at IBM learned to play better from its own games, as described in his 1959 paper [1]. Machine learning is part of artificial intelligence and overlaps with statistics and data science.
How does machine learning work?
- Gather data. Examples may be numbers in a table, pictures, text or sounds; see training data.
- Choose a model. The model is a mathematical function with adjustable settings called parameters.
- Train. The computer compares the model’s predictions with known answers and adjusts parameters to reduce errors, often using gradient descent.
- Test. The model is checked on data it has not seen to estimate how well it will perform in real use; see testing data and validation data.
- Deploy and monitor. The model is used in practice and watched for errors and changes in the world; see model drift.
This is how computers learn from data: not by understanding in a human way, but by finding statistical patterns that help them predict.
Types of machine learning
### Supervised learning
The training data include correct answers, called labels. A model learns to map inputs to outputs, such as email text to spam or not spam, or house features to price. Two main tasks are classification, which predicts categories, and regression, which predicts numbers; see supervised learning, classification and regression.
### Unsupervised learning
The data have no labels, and the model looks for structure, such as grouping customers with similar habits (clustering) or compressing data; see unsupervised learning.
### Supervised vs unsupervised learning
Supervised learning learns from labeled examples with a clear target, while unsupervised learning explores unlabeled data to find hidden patterns. Between them are semi-supervised learning, which uses a few labels with many unlabeled examples, and self-supervised learning, which creates its own labels from the data and underlies large language models.
### Reinforcement learning
An agent learns by trial and error, receiving rewards for good actions, as in game playing, robot control and tuning of chatbots; see reinforcement learning.
What is deep learning?
Deep learning is machine learning that uses neural networks with many layers. It excels with images, speech and language. The 2012 win of the AlexNet network in the ImageNet contest, reported by Krizhevsky, Sutskever and Hinton, is widely seen as the start of the modern deep learning boom [3]. See deep learning and neural networks.
Machine learning algorithms explained
- Linear and logistic regression: fit a line or a probability curve to data; simple and easy to interpret.
- Decision trees and random forests: ask a series of yes or no questions; forests combine many trees; see ensemble learning.
- Support vector machines: find boundaries that separate classes with the widest margin.
- k-nearest neighbors: classify by looking at the most similar known examples.
- k-means clustering: group data into clusters around centers.
- Naive Bayes: use probabilities for tasks such as text classification; see probabilistic AI.
- Neural networks: layers of simple units that learn complex patterns.
- Gradient boosting: build a strong model from many small corrections, popular for tables of data.
Machine learning vs artificial intelligence
AI is the broad goal of making machines act intelligently, while machine learning is one method of reaching it. Older AI used hand-written rules, as in expert systems. Today most successful AI uses machine learning. See artificial intelligence explained simply.
Machine learning examples and applications
- Email spam detection and text prediction.
- Recommendations on streaming and shopping sites; see recommendation systems.
- Speech recognition and translation.
- Image recognition and face unlock; see computer vision.
- Fraud detection and credit scoring; see AI in finance.
- Weather and demand forecasting.
- Predictive maintenance in factories; see AI in manufacturing.
Machine learning in healthcare, finance and education
Healthcare. Models help read X-rays and pathology slides, predict patient deterioration, support drug discovery and organize records; they need careful testing for fairness and safety; see AI in healthcare.
Finance. Banks and insurers use models for fraud detection, credit risk, trading signals and customer service, under regulatory oversight; see AI in finance and AI in banking.
Education. Systems adapt exercises to student progress, flag learners who may need help and automate parts of grading, with concerns about privacy and fairness; see AI in education.
How to train a machine learning model
- Define the problem and how success will be measured, for example accuracy or error.
- Collect and clean data, fixing errors and missing values and checking for bias; see data bias.
- Split the data into training, validation and testing sets.
- Pick a simple baseline model first, then try more advanced ones.
- Train and tune settings called hyperparameters, watching for overfitting, where a model memorizes the training data but fails on new data.
- Evaluate on the test set with suitable metrics such as precision and recall; see AI evaluation.
- Deploy and monitor, retraining when performance declines.
Beginners often use free tools such as Python with scikit-learn, or no-code platforms, to practice on small data sets.
Limitations
Machine learning needs good data, can inherit bias, may be hard to explain, can fail on unusual cases and sometimes finds shortcuts that do not hold up in the real world. See algorithmic bias and explainable AI.
Further reading and next steps
Readers can continue with deep learning, generative AI explained and AI ethics to see how machine learning connects with modern AI tools and their responsible use.
FAQs
What is machine learning in simple terms? It is teaching computers to learn patterns from data so they can make predictions without being given exact rules.
What are the three types of machine learning? Supervised, unsupervised and reinforcement learning.
Is machine learning the same as AI? No. It is a major approach within AI.
Do I need a lot of math to learn it? Basic algebra, statistics and programming are enough to begin, with more math helpful later.
How much data do I need? It depends on the task. Simple models can work with hundreds of examples, while deep networks may need millions.
More in Machine Learning
Key facts
- Arthur Samuel, who built a checkers-playing program at IBM, published a 1959 paper describing how a computer could be programmed to learn from experience and improve at the game1.
- Tom Mitchell's 1997 textbook defined learning as improving performance on a task with experience, a definition still widely quoted in courses2.
- The 2012 victory of the AlexNet neural network in the ImageNet image recognition contest, reported by Krizhevsky, Sutskever and Hinton, is widely seen as the start of the modern deep learning boom3.
Related concepts
Quick checkWhat is machine learning?Show answer
Machine learning is a branch of artificial intelligence in which computers learn patterns from data to make predictions or decisions, improving with experience instead of relying only on rules written by programmers.
Sources
- Samuel, Some Studies in Machine Learning Using the Game of Checkers, IBM Journal of Research and Development, https://ieeexplore.ieee.org/document/5392560
- Mitchell, Machine Learning, McGraw-Hill, https://www.cs.cmu.edu/~tom/mlbook.html
- Krizhevsky, Sutskever and Hinton, ImageNet Classification with Deep Convolutional Neural Networks, NeurIPS, https://papers.nips.cc/paper/2012/hash/c399862d3b9d6b76c8436e924a68c45b-Abstract.html
Editorially reviewed by Specialty Digest Editorial TeamLast reviewed November 17, 2026Researched and drafted with AI assistanceReport an issue
