Machine learning is the engine powering most of modern AI — from spam filters and product recommendations to language models and image generators. This machine learning glossary explains the most important ML terms in plain English, with beginner-friendly descriptions and practical examples for each one.
For a broader overview of AI terms, visit our main AI Glossary: 100+ Artificial Intelligence Terms Explained Simply.
What Is Machine Learning?
Machine learning is a subset of artificial intelligence where systems learn from data and improve over time without being explicitly programmed for each task. Instead of a developer writing rules for every situation, the model identifies patterns in data and builds its own rules. A spam filter that gets better the more emails it sees is a simple machine learning system.
Data Terms
Dataset — A structured collection of data used to train or test an ML model. A dataset might contain thousands of labelled emails (spam or not spam), thousands of labelled photos, or millions of rows of customer transaction data.
Training Set — The portion of a dataset the model learns from. The model sees this data and adjusts its internal settings to make better predictions.
Validation Set — A portion of data held back during training to monitor the model’s performance and help tune its settings, without being used to update its weights directly.
Test Set — Data the model has never seen, used only at the end to give a final, unbiased measure of how well the model generalises to new real-world data.
Label — The correct answer attached to a training example. For an image of a dog, the label is “dog.” Labels teach a supervised model what output to expect for a given input.
Feature — An individual measurable property used as input to a model. For a model predicting house prices, features might include square footage, number of bedrooms, location, and age of the property.
Types of Machine Learning
Supervised Learning — The most common type of machine learning, where the model trains on labeled input-output pairs. It learns to map inputs to the correct output. Example: training a model to detect credit card fraud using historical transaction data with fraud/no-fraud labels.
Unsupervised Learning — Training without labels. The model finds patterns, structures, or groupings in the data on its own. Used in customer segmentation, anomaly detection, and topic modelling.
Reinforcement Learning — A learning approach where an AI agent learns by taking actions in an environment and receiving rewards or penalties. Used to train game-playing AIs (like AlphaGo) and to align language models with human preferences (RLHF).
Classification — A supervised ML task that assigns inputs to discrete categories. Example: classifying X-ray images as showing a tumour or not, or labelling a customer review as positive, negative, or neutral.
Regression — A supervised ML task that predicts a continuous numerical value rather than a category. Example: predicting tomorrow’s stock price or a patient’s blood pressure reading.
Clustering — An unsupervised method that groups similar data points together without pre-set labels. Used in market segmentation, document grouping, and detecting unusual patterns in network traffic.
Model Building Terms
Model — The trained output of a machine learning process — a mathematical function that takes inputs and produces predictions. A trained ML model encodes everything the system learned from its training data.
Training — The process of feeding data to an ML algorithm so it can learn patterns and adjust its internal parameters to minimise prediction errors.
Hyperparameter — A configuration setting chosen before training begins that controls how the training process works. Learning rate, batch size, and number of layers are all hyperparameters. Unlike model parameters, they are set by the developer, not learned from data.
Overfitting — When a model learns the training data too precisely — including random noise and anomalies — and performs poorly on new data. An overfit model is like a student who memorises last year’s exam answers instead of understanding the subject.
Underfitting — When a model is too simple to capture meaningful patterns in the data, performing poorly on both training and new data. The opposite problem to overfitting.
Gradient Descent — The algorithm used to train neural networks by iteratively adjusting weights in the direction that reduces the model’s error. Think of it as repeatedly nudging the model toward better and better predictions.
Learning Rate — A hyperparameter that controls how large each step is in gradient descent. Too high and training becomes unstable; too low and training takes a very long time.
Neural Networks & Deep Learning
Neural Network — A computing architecture made up of layers of interconnected nodes (neurons), loosely inspired by the brain. Neural networks are the foundation of deep learning and power most modern AI applications.
Deep Learning — Machine learning that uses neural networks with many layers (hence “deep”). Deep learning models can learn highly complex patterns — like understanding speech, translating languages, or generating images.
Layer — A group of neurons in a neural network that transforms the input and passes it to the next group. An input layer receives raw data, hidden layers process it, and an output layer produces the prediction.
Weights — The numerical values inside a neural network that determine how much each connection between neurons matters. During training, weights are continuously adjusted to minimise errors.
Backpropagation — The process by which a neural network identifies which weights caused an error and adjusts them to reduce it, working backward from the output to the input layers.
Epoch — One complete pass through the entire training dataset. Training a model typically involves running many epochs to give it multiple chances to refine its understanding.
Batch Size — The number of training examples processed at once before the model updates its weights. Larger batches train faster; smaller batches can sometimes find better solutions.
Evaluation Metrics
Accuracy — The percentage of predictions the model gets right. Simple and intuitive, but it can be misleading when the dataset is imbalanced — for example, if 99% of emails are not spam, a model that always predicts “not spam” has 99% accuracy but is useless.
Precision — Of all the cases the model predicted as positive, how many were actually positive? High precision means few false alarms. Important when false positives are costly.
Recall — Of all the actual positive cases, how many did the model catch? High recall means few cases were missed. Important when missing a positive case is costly — like in medical screening.
F1 Score — A single metric combining precision and recall into one balanced number. Useful when both false positives and false negatives matter, and accuracy alone would be misleading.
| Metric | What It Measures | Best For |
|---|---|---|
| Accuracy | Overall correct predictions | Balanced datasets |
| Precision | Correctness of positive predictions | When false positives are costly |
| Recall | Coverage of actual positives | When missing cases is costly |
| F1 Score | Balance of precision and recall | Imbalanced datasets |
Frequently Asked Questions
What is the difference between machine learning and deep learning?
Machine learning is the broad category of systems that learn from data. Deep learning is a specific type of machine learning that uses neural networks with many layers. All deep learning is machine learning, but not all machine learning uses deep learning — simpler algorithms like decision trees and linear regression are also machine learning.
What does overfitting mean in practice?
An overfit model performs very well on its training data but poorly on new data because it has memorised the training examples rather than learned general patterns. You can detect it by comparing training accuracy (high) against validation or test accuracy (much lower). Common fixes include getting more data, simplifying the model, or adding regularisation.
Do I need to know maths to understand machine learning?
Not to use or apply it. Many ML tools and platforms abstract away the maths entirely — you can build and deploy ML models with tools like Google’s AutoML or Hugging Face without writing any equations. Understanding the maths more deeply helps if you want to build models from scratch or do research, but it’s not a requirement for practical use.
What is the difference between supervised and unsupervised learning?
In supervised learning, every training example has a label — a correct answer — and the model learns to predict that label. In unsupervised learning, there are no labels; the model finds structure in the data on its own, such as grouping similar items together. Supervised learning is used when you know what you want to predict; unsupervised learning is used to discover patterns you didn’t know were there.
Final Thoughts
Machine learning powers most of the AI tools people use every day, and understanding its vocabulary makes it much easier to follow developments in the field, evaluate AI products, and make better decisions when adopting AI for your business. The concepts are logical once you see how they connect.
For a complete AI vocabulary reference, visit our main AI Glossary. If you’re interested in how language models specifically work, our Generative AI Glossary covers the most relevant terms.
